Patent Application
20140226003
The present disclosure relates generally to charged particle beam (CPB) systems. More particularly, the present disclosure relates to imaging a material surface using a rastered beam system.
Examples of CPB systems include Scanning Electron Microscope (SEM) systems, Focused Ion Beam (FIB) systems and hybrids that include both CPB types, which are commonly known as “Dual Beam” or “Cross Beam” microscope systems. A Focused Ion Beam system is commonly referred to as a FIB. FIB systems produce a narrow, focused beam of charged particles, and scan this beam across a specimen in a raster fashion, similar to a cathode ray tube. Unlike the SEM, whose charged particles are negatively charged electrons, FIB systems use charged atoms, hereinafter referred to as ions, to produce their beams. These ions are, in general, positively charged. Note also that CPB systems may include multiple ion beams or multiple electron beams, perhaps in combination with each other.
These ion beams, when directed onto a sample, will eject charged particles, which include secondary electrons, secondary ions (i+ or i−) and neutral molecules and atoms from the exposed surface of the sample. By moving the beam across the sample and controlling various beam parameters such as beam current, spot size, pixel spacing, and dwell time, the FIB can be operated as an “atomic scale milling machine,” for selectively removing, or sputtering, materials wherever the beam is placed. The dose, or amount of ions striking the sample surface, is generally a function of the beam current, duration of scan, and the area scanned. The ejected particles can be sensed by detectors, and then by correlating this sensed data with the known beam position as the incident beam interacts with the sample, an image can be produced and displayed for the operator. The imaging capability of FIB systems, and of similar CPB systems, is advantageous for many applications where it is necessary or beneficial to analyze structures or features having nano scale sizes.
Dual beam system 10 also includes focused ion beam (FIB) system 11 which comprises an evacuated chamber having an upper neck portion 12 within which are located an ion source 14 and a focusing column 16 including extractor electrodes and an electrostatic optical system. The axis of focusing column 16 is tilted at an angle, such as 54 degrees from the axis of the electron column by example. The ion column 12 includes an ion source 14, an extraction electrode 15, a focusing element 17, deflection elements 20, and a focused ion beam 18. Ion beam 18 passes from ion source 14 through column 16 and between electrostatic deflection means schematically indicated at 20 toward sample 22, which comprises, for example, a semiconductor device positioned on movable stage 25 within lower chamber 26.
Stage 25 can preferably move in a horizontal plane (X and Y axes) and vertically (Z axis). Stage 25 can be tilted and rotated about the Z axis. A door or load lock 61 is opened for inserting sample 22 onto X-Y stage 25 and also for servicing an internal gas supply reservoir, if one is used. The door is interlocked so that it cannot be opened if the system is under vacuum.
An ion pump 28 is employed for evacuating neck portion 12. The chamber 26 is evacuated with turbomolecular and mechanical pumping system 30 under the control of vacuum controller 32. The vacuum system provides within chamber 26 a vacuum of between approximately 1×10-7 Torr and 5×10-4 Torr. If performing gas assisted processes such as etching or deposition, an etch retarding gas, or a deposition precursor gas is used, the chamber background pressure may rise, typically to about 1×10-5 Torr.
The high voltage power supply provides an appropriate acceleration voltage to electrodes in ion beam focusing column 16 for energizing and focusing ion beam 18. When it strikes sample 22, material is sputtered, that is physically ejected, from the sample. Alternatively, ion beam 18 can decompose a precursor gas to deposit a material on the surface of the sample.
High voltage power supply 34 is connected to liquid metal ion source 14 as well as to appropriate electrodes in ion beam focusing column 16 for forming an approximately 1 keV to 60 keV ion beam 18 and directing the same toward a sample. Deflection controller and amplifier 36, operated in accordance with a prescribed pattern provided by pattern generator 38, is coupled to deflection plates 20 whereby ion beam 18 may be controlled manually or automatically to trace out a corresponding pattern on the upper surface of sample 22. The liquid metal ion source 14 typically provides a metal ion beam of gallium. The source typically is capable of being focused into a sub one-tenth micrometer wide beam at sample 22 for either modifying the sample 22 by ion milling, enhanced etch, material deposition, or for the purpose of imaging the sample 22. Note that newer source technologies such as plasma, gas field ion sources and/or atomic level ion sources will produce other ionic species besides gallium.
A charged particle detector 240 used for detecting secondary ion or electron emission is connected to a video circuit 42 that supplies drive signals to video monitor 44 and receiving deflection signals from controller 19. The location of charged particle detector 40 within lower chamber 26 can vary in different configurations. For example, a charged particle detector 40 can be coaxial with the ion beam and include a hole for allowing the ion beam to pass. In other configurations, secondary particles can be collected through a final lens and then diverted off axis for collection.
A micromanipulator 47 can precisely move objects within the vacuum chamber. Micromanipulator 47 may include precision electric motors 48 positioned outside the vacuum chamber to provide X, Y, Z, and theta control of a portion 49 positioned within the vacuum chamber. The micromanipulator 47 can be fitted with different end effectors for manipulating small objects.
A gas delivery system 46 extends into lower chamber 26 for introducing and directing a gaseous vapor toward sample 22. For example, xenon difluoride can be delivered to enhance etching, or a metal organic compound can be delivered to deposit a metal.
A system controller 19 controls the operations of the various parts of dual beam system 10. Through system controller 19, an operator can control ion beam 18 or electron beam 43 to be scanned in a desired manner through commands entered into a conventional user interface (not shown).
In recent years, two and three dimensional imaging of large areas and volumes in a charged particle beam system such as SEM, FIB, or SEM/FIB combination microscope has attracted significant interest. Commercial systems such as the Carl Zeiss ATLAS two dimensional imaging system along with three dimensional imaging systems such as the FEI Company “Slice and View” along with methods described in U.S. Pat. No. 7,312,448 B2 have been available commercially. These techniques are generally performed on “bulk” samples, where the charge particle beam penetrates but does not transmit through the sample. It should be noted that this is quite different from the technique of electron tomography, which relies on the charged particle beam passing through the sample in transmission. While electron tomography is a well established technique in transmission electron microscopy, and can yield three dimensional datasets, these datasets are limited in scale due to the necessity of passing the electron beam completely through the sample and detecting it on the other side.
The aforementioned “ATLAS” two-dimensional and “Slice and View” style three dimensional techniques are sophisticated in their own right, however they both approach the problem of acquiring large datasets in a similar “step and repeat” fashion. In both cases a two-dimensional area is imaged either as a single image or as a collection of image “tiles” that may be “stitched” together to form a larger mosaic. Two-dimensional techniques tend to perform this step and repeat imaging over much larger areas than three-dimensional techniques, however three-dimensional techniques also remove a thin “slice” of material, then repeat the imaging process so as to build up a three dimensional dataset. This slice of material may be removed in several ways known in the art, including the use of a focused ion beam (typically at glancing angle, but occasionally closer to normal incidence), a broader ion beam which is often combined with some sort of mechanical beam stop to create a sharp edge, or an in-microscope ultramicrotome whose knife cuts away each slice.
CPB systems, such as FIBs or SEMs, have been used prevalently in the past for imaging small regions of a sample at high resolution. In the field of semiconductor circuits for example, typical structures being imaged include transistor devices and other small structures having dimensions from a few nanometers up to a few microns. In recent years, bio-medical applications are emerging in which higher resolution images for a large area of a sample are desired using the aforementioned 2D and 3D imaging techniques, and combining FIB and SEM. For example, imaging of a tissue sample having an area of 100×100 microns may be required in order to facilitate visual identification of a particular structure of interest which may be present. Accordingly, a high resolution image of the entire area is required, otherwise it may not be possible to visually identify the structure of interest. Furthermore, the particular structure of interest may lie within a plane different from the exposed area being imaged. In this example, if the imaged area of the sample is defined by an x-y plane, then the tissue sample has a depth component, z. Therefore sections of the tissue sample are taken at predetermined depths and the newly exposed area is imaged.
The problem with currently known techniques is the large amount of time required to image large volume samples at high resolution. The increasing demand for 3D high resolution images of 100 μm×100 μm×100 μm volume samples is problematic. Typically sections are prepared ˜15 μm Sections @ ˜3 nm pixels with ˜9 nm depth per slice using a FIB; more typically three times this depth per slice is all that can be achieved using other sectioning methods which can be used. Typical dwell times for the electron beam are on the order of 1 μs per pixel in order to obtain sufficient signal to noise. At 3 nm voxels with a dwell point time of 1 μs, 20 minutes of imaging time alone per section are required, and about 110 hours per μm of depth sectioned, which must be multiplied by 100 to section through 100 um of depth, and this is imaging time alone, i.e. it is assumed sectioning occurs concurrently or near instantly. Therefore a total of about 1.5 years of time is required to image a 1,000,000 μm3 cube, assuming that the CPB system is capable of operating for this continuous period of time without malfunction or interruption, or the sample undergoing sectioning can be reacquired and realigned in an acceptable manner. Another issue related to imaging large areas is the fact that the sample is vulnerable to “drifting” during the imaging process, in which the sample moves due to mechanical variations in the stage supporting the sample, and/or thermal effects on the environment of the microscope.
It is, therefore, desirable to provide a method and system for reducing the amount of time required for CPB imaging while maintaining accuracy.
It is an object of the present disclosure to obviate or mitigate at least one disadvantage of previous CPB systems.
In a first aspect, there is provided a selective high resolution imaging method for a charged particle beam apparatus. The method includes acquiring and displaying a sample area image of a sample at a first resolution; scanning at least one exact region of interest in the sample area image; and acquiring and displaying an image of the at least one exact region of interest at a second resolution greater than the first resolution.
According to the embodiments of the present aspect, the sample can be cross-sectioned to expose a new surface, of which the same exact region of interest is imaged at the second resolution. This sequence of sectioning and imaging the exact region of interest can continue until a new sample area image of the sample at the first resolution is requested. At this time, new exact regions of interest can be added, or the previous exact region of interest can be modified.
In yet further embodiments, alignment vernier notches can be formed on the sample, which are visible as a pair of objects in cross-section that approach each other in distance as further cross-sections of the sample are taken. Comparisons of the notch distances from a current to previous cross-section can be used to determine the exact cross-section thickness, for the purposes of adjusting a milling rate of the FIB.
According to further embodiments, any beam of the charged particle beam apparatus can be controlled with accuracy by use of a multi-digital to analog circuit, which receives a primary digital code corresponding to a primary deflection voltage for moving a position of the focused ion beam, and at least one additional digital code proximate to the primary digital code for generating a secondary deflection voltage. These deflection voltages are averaged to provide a final deflection voltage.
Other aspects and features of the present disclosure will become apparent to those ordinarily skilled in the art upon review of the following description of specific embodiments in conjunction with the accompanying figures.
Embodiments of the present disclosure will now be described, by way of example only, with reference to the attached Figures.
Generally, the present disclosure provides a method and system for improving imaging efficiency for CPB systems while maintaining or improving imaging accuracy over prior CPB systems. According to the present embodiments, a large field of view image of a sample is acquired at a low resolution and thus, at high speed. The low resolution level is selected to be sufficient for an operator to visually identify structures or areas of interest on the low resolution image. The operator (or an image analysis subsystem, which will be considered another type of “operator”) can select one or more small areas of arbitrary shape and size on the low resolution image, referred to as an exact region of interest (XROI). The outline of the XROI is mapped to an x-y coordinate system of the image, and the CPB system is then controlled to acquire a high resolution image of only the XROI identified on the low resolution image. For 3D imaging, once the XROI is identified, each section of the sample can be iteratively imaged in the previously described manner, with the operator having the option to redefine the XROI later. The operator may also observe the information contained in the XROI image data and redefine the XROI based on this information.
Many CPB systems have an accessible console with removable boards inserted therein to control particular functions of the CPB system 10. Preferably, CPB workstation 100 includes a microprocessor, memory and mass storage, typically embodied as a computer workstation with a monitor, and a CPB system hardware interface 102 which can be connected to the system controller 19 of the CPB system 10. In the presently shown embodiment of
The presently described embodiments are executed by the CPB workstation 100 for controlling the CPB system 10 components to improve imaging throughput while maintaining image quality. In a first embodiment, referred to as a selective high resolution imaging method, only specific regions of interest on a 2D large field of view image of a sample are acquired by the CPB system 10. Therefore, significant time savings are obtained because the entire large field of view is not imaged at high resolution. In the present embodiments, any number of specific regions of interest, referred to as exact regions of interest (XROI) can be inputted to the CPB workstation 100 for acquiring high resolution images thereof, or a series of image resolution levels from low to high resolution.
This low resolution XFOV image 202 is displayed for the operator, who can analyze the image and determine the presence of one or more desirable XROI. At 204, the operator can create an outline of any arbitrary shape on image 202, indicating a desired XROI. This is shown in
To enhance the context of the higher resolution images 212, the CPB workstation 100 can overlay the higher resolution images 212 over the low resolution XFOV image 202 so that the operator can generally view the structures surrounding the higher resolution image areas. An example of this is shown in
Accordingly, it is not necessary to image the entire XFOV image 202 at high resolution, as this would consume a significant amount of time when the area of interest may only occupy a small portion of the image. Therefore, the combination of acquiring a low resolution image of the large XFOV area at high speed followed by selective high resolution image acquisition of smaller XROI areas improves can significantly improve the overall imaging throughput of the XROI at high resolution, relative to the prior art schemes. The final images can be subjected to graphical post-processing, such as for example by adding virtual colouring to features having the same particular grey-shading intensities.
Another factor for improving overall imaging throughput is the ability to select XROI outlines of shapes which can approximate the outline of an area of interest. As shown in the previous example images, choosing a precise XROI, rather than being limited to merely scanning a rectangle that contains the region of interest, allows a reduction in the number of pixels to be scanned, thus increasing throughput. Any other improvements to the system, such as signal to noise, detector efficiency, beam current and spot size improvements, merely make the XROI approach more efficient. For example, a common yeast cell is approximately spherical. It is known that the volume of a cube of diameter D is approximately twice the volume of a sphere of diameter D. Thus, if one is constrained to imaging a contant rectangular or square area, the CPB system requires approximately twice as long to image relative to imaging a circular XROI. Equations 1 and 2 below mathematically illustrate this relationship.
V
Cube
=D
3 Equation (1)
V
Sphere= 4/3πr3≅½D3 Equation (2)
The previously discussed selective high resolution imaging method throughput benefits are significantly scaled when applied to generating data for 3D reconstruction of a sample. This is due to the fact that for each new section of the sample, a new image of a region of interest taken at high resolution is acquired. As previously discussed, the prior art technique of obtaining high resolution images of a large area XFOV for a plurality of sections can be unacceptably long.
Now that the region of interest of the current section of the sample has been imaged at high resolution, the method proceeds to 308 where a new section of the sample is obtained. In
In the present example of
By periodically acquiring “Key Frame” images of the entire sectioned surface of interest, this allows the automatic or manual selection of new XROI's that may appear during sectioning, and the dynamic changing of the existing XROIs. Note that the image data from each (typically higher resolution XROI) can also be interrogated to determine if, on the next or future XROI imaging passes, the boundaries of the XROI should be modified.
According to a further embodiment, of the present invention, overall imaging throughput can be increased by examining a particular region of interest and determining based on pixel intensity of a rapidly acquired lower resolution image, whether an area should be scanned. For example, when imaging biological material stained using common protocols that introduce heavy metals into the tissue, and observing/detecting backscattered electrons with inverted contrast (as is common), tissue appears dark (more signal, resulting in a dark pixel when observing with inverted contrast), and embedding material appears brighter (less signal, resulting in a bright pixel when observing with inverted contrast). These images can be processed (including Key Frames or XROIs) to detect regions containing only embedding material, and avoid imaging these regions of embedding material, or avoid imaging regions of embedding material that are sufficiently distant from regions that are identified as not being embedding material (thus imaging regions of embedding material near what may be true sample), so as to increase throughput.
Alternately, it is possible to analyse the signal as it is being accumulated within a single pixel or dwell point, and determine during a subsampling time, such as the initial subsampling time, for that pixel whether sufficient signal has been detected to expect that pixel is a region of embedding material, and not a region of tissue. If it is determined that the pixel is tissue, the dwell continues to improve signal to noise; if not, the beam is advanced to the next pixel before the full dwell time is reached, thereby improving throughput. The “advanced from” pixel (presumed to be embedding material) may have it's intensity normalized as if the signal had been acquired for the full dwell time, and it may also be flagged somehow to indicate that it was not dwelled for the complete period, and optionally how long a dwell time did occur. Additionally, the dwell time required to achieve a certain number of counts can be recorded, advancing to the next pixel when a predetermined number of counts has occurred, and thus generating a “dwell time to a given number of events” image map, rather than an image map of the intensity observed in a fixed dwell time.
According to another embodiment, the CPB system 100 can be configured to analyse information from clusters of neighbouring pixels to determine if an “advanced from” pixel should truly have been advanced from, or if perhaps it's signal did not occur during the subsampling time do to some anomaly, based the known methods of Poisson and other discrete particle counting statistics. Thus a particular pixel within the same image can be revisted/re-imaged based on processing of neighbouring pixels should those pixels appear to indicate a high probability that a particular pixel has been undersampled. In systems where slices are archived, a neighboring pixel can be within other slices, thus the other slices can be re-imaged at a later time.
The “time of flight” between a charged particle leaving the CPB column, impacting the sample to generate a secondary signal, and this secondary signal being detected can be significant enough that a considerable latency can exist in the system, in which case an image can be accumulated from multiple “subimage” passes over the same physical region, each pass taking sufficient time that the individual pixel latency is small in comparison to the time per pass to acquire a subimage. The exact scan strategy to create the next subimage may be modified based on an analysis of pixel information in one or more subimages, using the methods described above, ultimately building up a final image potentially composed of pixels that have seen various total dwell times, in a similar manner to the fashion described above, but overcoming the impact of time of flight latency or other latencies such as detector response or dead time. Note also that image alignment techniques may be used between subimage passes to correct for sample or instrument drift which would otherwise cause the system to perform subimaging passed in what is potentially the wrong location.
As the end data set of this serial sectioning and imaging method ultimately ends up with three dimensions of data, represented by a “stack” of images which may be aligned, there are cases where one can infer from a pixel in stack N whether a pixel at stack N+1 requires imaging. Thus, referring to an earlier example, knowing no tissue is present in a pixel in stack N, it may be sufficient not to bother imaging that same position in stack N+1 (or indeed up to N+M). Alternately, reconstruction algorithms may be able to interpolate or otherwise deduce the necessary data even from “sparse” images, for example if only half of the pixels are imaged on even sections, the other half on odd slices (after the fashion of a checkerboard—“red squares”=pixels imaged on even slices, “black squares”=pixels imaged on odd slices. Image processing at the acquisition level may be used to interpolate or otherwise “fill in” the missing data to increase throughput. Note that this is not limited to a “checkerboard” approach, and as the rate limiting step is often the SEM imaging, these “sparse imaging” approaches, be they temporal or geometric can lead to greater throughput which can be used to achieve finer slice resolution in the FIB/SEM system, which in turn may allow for more intelligent algorithms to supply the intensity values for those pixels “skipped” using sparse methods.
The previously described 2D and 3D imaging methods can benefit from additional improvements over control of the CPB system 10, which can be provided by CPB workstation 100. These are referred to as multi-pass rastering, spatial super-sampling and temporal sub-sampling, which can be optionally enabled during the imaging phase in order to improve data quality or optimize a particular component of the signal that is used to generate the image.
While the previously described 3D imaging method is described for a dual beam FIB-SEM system, the techniques can be applied to imaging methods where the slices of a sample are archived, such as by using the “ATLUM/ATUM” technique developed by Dr. Jeff Lichtman et al. of the Department of Molecular and Cellular Biology of Harvard University. In the Lichtman technique, slices of a sample are pre-prepared and subsequently imaged. Therefore an operator can return to any slice for re-imaging any particular region of interest. The application of the present 3D imaging embodiments to archived slices includes performing imaging at multiple resolutions—a first, lower resolution pass through a series of sections, which are then image processed to determine where higher resolution imaging is to occur, after which further, higher resolution images are acquired. The previously described imaging methods can be applied to all stages of this process.
According to further embodiments, the previously described embodiments can complement the technique of Lichtman, which may be used to acquire sections that are relatively thick—for example, 300 nm in thickness. For example, “lower resolution” imaging of each section and its XROIs (which are not necessarily limited in area and may indeed be the entire section) is executed by the CPB system 10. Further processing then determines regions that require higher resolution data acquisition. At this point, it is possible to use the FIB/SEM approach to slice through the “thick” section in a serial manner, for example, sectioning at 90 degrees to the originally sectioned surface, and obtaining a higher resolution data set within this FIB/SEM sectioning area on each desired thick section, once again applying the XROI technique as desired to further improve throughput.
During the process of imaging one or more XROI's, it is possible that a given XROI or the agglomerate of XROI's desired may exceed the maximum size of image unencumbered by differential non-linearity (DNL) artifacts that is available using the digital to analog (DAC) hardware. This can be dealt with using techniques such as the multi-DAC approach discussed later, and/or through the use of a mosaic tiling approach, sometimes referred to as montaging of an image. The current state of the art requires the use of rectangular image tiles, however as discussed below, this can be improved in many cases using non-rectangular tiles.
In order to image a large area at high resolution, it is necessary to acquire a mosaic of multiple smaller images at high resolution and stitch them together. This is commonly done with square or rectangular images as they are normally obtained from a scan generator. However, as the size of the field of view becomes large for ultra-high pixel density images, there may be scan and beam distortions that limit the extent of the images. Examples of scan distortions include a “pin cushion” type distortion where the normally straight edges of a square or rectangle appear deflated, and a “barrel” type distortion where the normally straight edges of a square or rectangle appear inflated. Examples of beam distortions includes focus and astigmatism distortions. These types of distortions are well known in the art.
Since these distortions typically have circular symmetry, the greatest distortions occur in the corners of a square or rectangular image. To avoid these artifacts yet still use the largest field of view possible to minimize the number of images required, the CPB system 10 can be configured by the CPB workstation 100 to acquire images with a hexagonal shape that tile to completely fill the mosaic while allowing for efficient stitching. Therefore, overall image quality is improved over rectangular tiling, while improving image capture throughput. The mathematical reasoning behind using hexagonal shaped tiles is further explained.
If it is assumed that the distortions become unacceptable a distance r from the centre of the field of view, then the area of the largest hexagon with acceptable distortions is 3√{square root over (3)}/2r2≈2.59r2, compared to 2r2 for a square image. This means 30% fewer images are required when using hexagons compared to squares, which are the most efficient rectangles, thereby requiring fewer stage movements and less stitching. For comparison, using a non-square image with an aspect ratio of α, the area is even smaller than for a square image (A=4r2/(α+1/α), 0≦A≦2r2). Any other combination of shapes that completely fills the space can be used, such as octagons and diamonds by example.
Creating such a mosaic requires some translation from tile to tile, be it using a Pan/Shift capability of the microscope, or for larger distances a physical motion of the microscope stage. In the case of a physical motion, in either 2D or 3D image acquisition, a technique of high resolution XROI imaging can be employed to improve the accuracy of the stage motion as follows.
In many instances, the stage should be moved by a very precise amount relative to the current position. If the stage is not accurate enough, due to mechanical limitations for example, then such a move is not normally possible to the required precision. Assume for example that one wants to locate a specific contact in a semiconductor memory array. Because of the repetitive nature of the array, it is only possible to identify a contact based on its exact position relative to the corner of a memory array which is unique enough to be identified. Using a stage movement from the corner results in a given uncertainty because of the inaccuracy and imprecision of the stage. If the cell size is smaller than the uncertainty of the stage move, then it is not possible to identify which cell and which contact around the resulting approximate position is the actual target. According to a present embodiment, the stage can be moved by the desired amount, followed by an identification of precisely how much it has actually moved. Any additional correction is then accomplished by precisely shifting the beam. This is achieved by using high resolution imaging at a large field of view (FOV) and pattern recognition.
The method of
According to an alternate embodiment, the FOV can be only as large as the required stage motion by doing two stage moves and reacquiring a reference image between the two. Assuming a high resolution image at 32768 pixels along the FOV and an absolute scan accuracy of 4 pixels, this method would result in a precision of 10 nm for stage movements up to 80 microns, or 25 nm precision is achievable for stage movements up to 200 microns. In any implementation, an improvement in scan accuracy will directly result in more precision or larger stage movement given the same precision requirements.
In another alternate embodiment, the method involves continuously scanning a region of interest at a frequency sufficient that any motion in that feature would be a small increment during a single image pass, and tracking that feature across a high resolution scan (this is greatly enable by a high resolution of 32k×32k or higher existing technology being typically limited to 8k×8k) while the stage is moved to the desired location—thus by relying on the well calibrated scan and tracking the feature continuously through high speed scanning, a precise determination of the position of the feature can be achieved, relative to the accuracy of the scanned image field, throughout the entire period during which the stage is moving, which typically will yield greater positional accuracy than can be achieved by a mechanical stage system.
When acquiring images at large fields of view, it is common to encounter distortions that affect the quality of the image. These may be divided in to three general groups: scan distortions due to lensing or sample tilt effects, beam distortions such as astigmatism and focus differences across the image or sample induced distortions. According to present embodiments, the CPB workstation 100 is configured to mitigate or eliminate some of these distortions through active and/or passive processes. Active distortion mitigation methods include modifying the scan and beam conditions. Passive distortion mitigation methods include post-processing of the images and some correction of these artifacts. Following is a non-exhaustive listing of possible distortions that could be corrected for with the presently described distortion mitigation embodiments: (i) loss of focus and stigmation, particularly in the corners of the image; (ii) barrel or pincushion distortion at large FOV; (iii) leading edge distortion due to beam dynamics; (iv) tilt parallax (trapezoidal distortion); (v) leading edge distortions due to minor sample charging; and (vi) focus changes due to sample geometry.
One embodiment for large area distortion mitigation is dynamic scan and beam compensation.
When dealing with geometric distortions, with the use of a digital scan generator, it is possible to apply a correction to the scan such that the actual scan after distortion by the column and/or sample produces the original desired result. Consider the microscope has a non-ideal transfer function (
) that converts an input scan position
into a real position
If the inverse T−1() of this transfer function is known, then the desired scan position is obtained when a corrected input
is fed to the system:
Accordingly, the correction can be generated by two different techniques according to the present embodiments. In one embodiment, the inverse of the transfer function is parametrized analytically and the parameters are adjusted until the proper output is obtained. In another embodiment, a calibration grid is used and a map of the inverse transfer function is built numerically based on the measured discrepancy between the input and the output. In either case, for a digital scan generator equipped with a digital signal processor (DSP) or a field programmable gate array (FPGA), the corrected input can be calculated or accessed from a lookup table and applied directly in the scan generator hardware. For simple corrections such as a tilt parallax, the function can be implemented as an analytical function, but for more complicated corrections, a predetermined lookup map can be employed.
A second embodiment for large area distortion mitigation is dynamic focus tracking along a cross-section.
In the situation when imaging a sample that is tilted and two faces are exposed to the beam (for example the original surface of the sample and the cross-sectioning face), the focus of the beam is dynamically adjusted as a function of the position in the image in order to keep in focus along both faces of the sample. Current implementations of dynamic focus are limited to allow tracking within one plane, either the surface or the cross-section. By having a more complex tracking routine, it is possible to determine, based on the current scanning position in the image, which plane is being scanned and thus adapt the focus appropriately. The focus is constantly adjusted according to the known sample topography to preserve focus on the entire area, provided the focus adjustment can be performed sufficiently fast to keep up with the scan. Under these conditions, the focus can be adjusted in order to mitigate defocussing effects away from the centre of the image by constantly adjusting the focus according to a predefined map. If optimal stigmation varies within the image field of view, then it is also possible to adjust the stigmation according to a pre-established map, according to the position of the beam in the image.
Currently, calls to adjust the focus and stigmation are generated in software by the native CPB system 10, so changes to the focus and stigmation within a scan line are only possible for very slow scans. According to the present embodiment, the scan generator is configured to output not just the x and y deflection signals but also a focus and stigmation correction signal, which would make the system operational for regular scan speeds. In this embodiment, the scan generator is configured to include a lookup table in memory for both the focus and stigmation as a function of beam position. As the scan is generated, the focus and stigmation outputs are converted to signals usable by the microscope using a standard digital to analog conversion and amplification mechanism.
Either in combination with or independently from the dynamic beam and scan and focus tracking embodiments above, the images can be post-processed to compensate for large area imaging distortion.
Assuming an image was acquired with a known distortion, post-processing operations can be performed on the image to remove the effect of the distortion. This is commonly done in optical imaging or photography where some lens artifacts are well defined. Wide angle lenses tend to have some degree of barrel distortion which is commonly post-corrected inside the camera as the image is saved to a file. According to the present embodiment, the CPB workstation 100 is configured to include the same type of process, where the acquisition engine automatically morphs the image as it is acquired so that the output obtained by the operator is free from distortions. As in the case of in-camera processing, the distortion should be well established prior to acquisition.
One example implementation is for imaging moderately charging samples. When imaging a charging sample with an SEM, the act of scanning the beam results in charge accumulation on the sample surface which slightly affects the beam position. Under certain conditions, the scanned area of the image will have a systematic compression on all lines, on the edge where the scanning begins. In this case a standard raster scan is used, imaging all lines from left to right. By comparing the resulting image to the known geometry of the sample, a simple model of exponential decay can be used to accurately model the amount of lateral shift of each pixel from its nominal position,
(where A is the shift of the left-most pixel and τ characterizes how fast the shift decays to 0). Using this model, the necessary warping of the measured image can be performed in real-time such that the operator is shown a proper image free of the artifact.
Another contributor to large area imaging distortion is dynamic drift. Acquisition of a high pixel density image takes much longer than under conventional imaging conditions, and therefore the stability of the microscope may result in unacceptable drift during the duration of the acquisition. The sources of this drift may be electronic (beam deflection shift) or mechanical (stage drift). Typically, the largest source of drift is stage drift, assuming the microscope and electronics are up to stable operating conditions.
According to a dynamic drift compensation embodiment, this drift is compensated for by shifting the beam systematically at the scan level according to a predictive drift model. This model is developed from prior images, where, for example, a given system has a known relaxation drift after a stage move. Alternately, the drift model can be generated dynamically by regularly pausing the imaging and performing registration on a fiducial to evaluate the current amount of drift. A model can then be applied to anticipate the amount of correction that is needed to compensate for the drift.
As an example, a 32k×32k image of 1 Gigapixel is acquired using a dwell time of 2 μs will take approximately 35 minutes. Assuming the system stage drift specification is 3 nm/minute, the stage may have drifted by nearly 100 nm at the completion of the image. If the image was acquired with a resolution of 5 nm, this will result in an error of 20 pixels between the top and the bottom of the image. By pausing the image and determining the amount of drift periodically, this error can be reduced. By way of example, by pausing every 5 minutes, it is possible to reduce the error to 15 nm (3 pixels), or less if the drift is systematic and can be modeled properly.
According to the present embodiments, it is also possible to measure the change in environmental variables such as for example, the temperature, sound, vibration or pressure, in close proximity to the sample or the microscope, and by evaluating the drift as a function of such environmental variables, create a predictive model for the impact a change in such a variable may have on the image (including drift). For example, it may be that a rise in temperature typically corresponds to a certain drift in a certain direction that lags the measured temperature rise by a certain time. Thus it may be possible to dynamically adapt the scan to (optimally smoothly) compensate for this drift during the course of acquisition of one or more images.
It is also possible to look for rapid changes in these variables, such as the sound of a slamming door, and (either additionally or in place of) also evaluate the local portion of the image as acquired (such as the last few scan lines) against the average metrics of the image. At present, this can be done by analyzing the entire image on completion. According to a present embodiment an algorithm can be developed to analyze pairs of lines in the image for the purposes of calculating the following:
However, the current art requires completion of the image, and post-calculation of the PASS or FAIL state, which can in turn trigger the entire image to be reacquired. It is advantageous to look for problems with the image either by dynamically performing such analysis as the image is acquired, or by measuring environmental variables and looking for signatures of events previously determined to cause problems in the image. When a likely problem is discovered, for example a sound is detected by an acoustic monitoring circuit that is above the threshold determined to be sufficient to cause a deficiency in the image it is then possible to repeat a portion of the image to remove the problem. For example, if a vibration is detected either by analyzing the lines of the image as discussed above, or by observing an environmental event (such as a sound spike caused by the slamming of a door), the acquisition can be stopped and the beam “backed up” to a point spatially that had been scanned prior to the event being detected, and this small portion of the image can be re-scanned, rather than requiring the entire image to be rescanned. This also has the advantage that the “back up and repeat” happens temporally very close to the original scan, leaving little time for errors such as drift to occur. And additional step of “drift correction” to align the last “known good” portion of the image prior to the environmental event and the first portion of the backed-up and re-imaged portion may be performed during acquisition to ensure the final image that is saved has a seamless transition across the portion of the sample that was being imaged when the original environmental event occurred.
In addition to automatically attempting to improve the imaging conditions of the microscope, the operator may wish to have more direct control over the imaging conditions.
When acquiring very large images, it is difficult to determine if imaging conditions such as focus, stigmation, etc. are optimized for the best trade-off across the entire image. It is therefore advantageous to be able to have a single display that allows the operator to see multiple regions of the large image at full, or otherwise high, resolution at essentially the same time. According to a present imaging condition optimization embodiment, the operator can view multiple images taken at different areas of the sample simultaneously for the purposes of manually adjusting imaging conditions to obtain the optimal trade-off for best results in all parts of the image.
Window 570 is referred to as a multi-region image, which now allows the operator to adjust the focus, stigmation, beam shift, and other CPB conditions to obtain the optimal trade-off for best results in all parts of the image. According to an aspect of the present method, the five regions scanned have the same area (number of pixels), so each requires the same time to scan. In alternate embodiments, more than 5 regions of the ROI can be scanned and displayed in window 570. For example, a 3×3 matrix showing the center of edges plus the four corners plus the center is possible. Furthermore, the operator can reposition these regions within window 570 as desired, i.e. it is not required that the actual regions scanned have exact positional correspondence to the four corners and the center, nor that each region scanned have the same area as the other regions.
It is also desirable to apply local image processing to adjust for stigmation or focus issues within different regions of the image once the best results are obtained through optimizing CPB conditions. Such local processing can be applied to each of the multiple regions described above, then the determined values interpolated between regions. Determining optimal local settings may be done by a human user, or through image processing techniques.
When two charged particle beams are used for 3D analysis involving both material removal slice-by-slice and imaging, as shown in
Maintaining knowledge of the position of the cross-section face in three dimensional space is vital to ensuring each slice is at or close to the desired thickness, and also deriving a knowledge of the actual thickness of each slice. A 3D positional tracking method is described that enables the tracking of the position of the cross-section face during sectioning and imaging in such a way that a direct measurement of the actual position of each slice can be calculated, allowing for dynamic repositioning to account for drift in the plane of the sample and also dynamic adjustment of the forward advancement rate of the FIB to account for variations in the sample, microscope, microscope environment, etc. that contributes to drift. An additional result of this approach is the ability to dynamically calculate (and potentially report to a dynamic image processing module) the actual thickness of each acquired slice as it is acquired.
Note that it is also possible to nanofabricate the “notches” directly on the surface of the sample 600 (in the absence of a first protective layer), and it is also possible to fabricate the “notch” as a raised structure rather than a groove, i.e. depositing rather than removing material. The term “notch” is understood to refer to a structure deliberately nanofabricated for purposes of alignment, such as a line or curve, whose geometry is known. It is often desirable that this structure be contiguous, however noncontiguous structures such as a dotted line may also be employed, and additional information may be gleaned from the “duty cycle” of the “dots”. Multiple dotted lines with different or offset duty cycles may be employed.
In a similar fashion, one or more parallel or nearer parallel “notches” 618 may be nanofabricated to serve as known good targets for autofunctions such as auto focus, auto stigmation, auto brightness, auto contrast, etc. as these features can be fabricated to have a known and constant position on the cross-section face during all or a portion of the cross-sectioning process. In the present embodiment shown in
From a “top down” perspective, a number of notches are nanofabricated onto the surface of the first protective layer. These notches converge at a predefined angle and a set of them may have the appearance of chevrons, although it is not a requirement that the notches meet at a point nor have any other specific geometry relationship other than the fact they are not parallel and their geometry is known. In
The cross-section image of
As is readily seen in cross-section, the sample has a surface which has a certain surface roughness. Depositing a first protective layer over the surface, could have the additional benefit of providing a degree of planarization, smoothing out a portion of the roughness.
In the microscope, it is also more readily observed in cross-section that it is desirable that there be contrast between the first protective layer and the second protective layer. One way this can be achieved is if the average atomic number of the material from the first protective layer is sufficiently different from the average atomic number of the second protective layer. This can be accomplished by depositing one of the layers using a heavier (higher average atomic number) materials such as deposited “platinum” or “tungsten” from a precursor gas such as tungsten hexacarbonyl (W(CO)6). Those skilled in the art will realize that the process of ion (such as Ga+ or He+) or electron beam deposition from a precursor gas is well known, and also leads to a “tungsten” deposition that incorporates a mixture of W, C, and the incident beam (Ga, etc.). A lighter (lower average atomic number) material such as “carbon” or “silicon oxide” can be deposited for the other layer. When viewed using a detector sensitive to the average atomic number (i.e. one such as the Carl Zeiss Energy Selective Backscatter Detector, EsB), regions of higher average atomic number have higher signal (brighter) and regions of lower average atomic number have lower signal (darker). Note also that the EsB allows imaging of SEM generated electrons of a certain energy and filters out the FIB generated electrons during simultaneous milling and imaging.
It is also possible to achieve the desired contrast between the first and second layers using a single gas precursor, and depositing one layer using a first beam (say an ion beam such as Ga+, He+, Ne+ or Ar+ by example) and a second layer using a different beam (an ion beam of a different species or an electron beam). In the case of using a Ga beam for one layer and an electron beam for another, the average atomic number of the two layers would be different due to factors such as the incorporation of the Ga into one layer (whereas the deposition by an electron beam would not leave an elemental species incorporated in the layer), differences in density of the layer due to different chemical processes arising from the deposition method, etc.
It is also possible that the one or both of the protective layers is omitted, and the contrast arises between the features nanofabricated into or onto the sample and the sample itself (and any protective layers thus employed).
In the presently disclosed embodiments, the notches are used as alignment marks, ie. patterns in the sample that are such that when imaging the cross-section face, the distance between the marks allow unique identification of the position of the cross-section plane along the Z axis. These alignment marks can be repeating structures that only allow unique identification of their position when a coarse position of the cross-section is also known, and can be patterned directly into the sample surface using the patterning beam. In one example, the notches are patterned such as to produce a suitable contrast when imaging the cross-section. For example, the alignment mark generated by first depositing a Pt or W layer on the sample surface, milling the marks and the depositing a C layer on top. This arrangement results in a high contrast image with most imaging beams and detectors. The second protective layer can be a material with high contrast relative to the material of the notches, to further enhance notch patter recognition by the CPB workstation 100, since the combination of the contrasting layers would be unique in the image, and thus easily detectable by the system for auto depth calculations. This feature is now described in more detail.
Referring to
Where these notches intersect the XY plane at z=zCS, the distance between them is
If this distance is measured at some time taken as the origin and a time t later, then the precise position of the cross-section time can be determined from the change in distance between the notches:
If the notches are patterned at ±45°, then
and this simplifies to:
The change in distance between the notches is twice the change in z of the cross-section position. This means that under these conditions, with a precision of 2 nm when measuring the distance between the notches, the position of the cross-section can be determined to within 1 nm. According to the present embodiments, the method of
In
According to the present embodiments, the notches are imaged in the cross section view and can be used by the CPB workstation 100 to execute other different operations. The CPB workstation 100 can calculate the position of the mill based on the alignment marks, compared to where it is expected to be based on the intended position of the milling beam. Therefore, it is possible to determine the amount of drift of the sample relative to the milling beam. The source of this drift, be it stage, sample, beam electronics, etc. is unimportant, as it simply relates to the actual versus estimated position of the milling beam relative to the sample. A suitable model can be used to project future drift based on observed values and preemptively adjust the milling beam position to match a target milling rate.
In other application of the notches, the position of the sample surface in the X and Y plane (cross-section imaging position) can be determined. By calculating shifts in the image based on the notches, it is possible to determine the amount of drift of the sample relative to the imaging beam. The source of this drift is unimportant, it simply relates to the actual versus estimated position of the milling beam relative to the sample. A suitable model can be used to project future drift based on observed values and preemptively moderate the imaging beam position to minimize any drift that may occur during the image acquisition.
In yet another application, the CPB workstation 100 can normalize the intensity of the image based on the light and dark portions of the protective layers above and below the notches. When the notches are created as a stack of two materials of different CPB imaging contrast, the histogram of an image of the bilayer above and below the notches will generally be bimodal. The average and spread of the two modes of the histogram can be used to evaluate or compensate for brightness and contrast changes of the detector or the beam current, etc. of the CPB beam itself.
Additional applications for the notches include:
1) Automatic sample realignment. Using multiple fiducials would allow full and precise repositioning of the sample in the event the stage was moved or the sample unloaded.
2) Automated aperture alignment, in the event multiple milling apertures need to be used.
3) Automatic realignment in the event of a glitch (or power off/power on) of the milling beam.
4) Autofocus and autostigmation of the milling beam.
It is advantageous to correct for any drift in a predictive manner, allowing for a smooth, adaptable adjustment of both the milling and imaging beam position (or adjustment of just one beam) to predictively correct for any error measured from the fiducial marks discussed above, commonly referred to as “drift”. It is often optimal that this smooth correction for drift be performed as a series of small corrections separated in time, rather than larger corrections that are more discontinuous in nature.
A slice of the sample is milled away by rastering the milling beam according to a pattern that is predominantly perpendicular to the thickness of the slice to be removed. In its simplest implementation, the milling beam is rastered in a single line, perpendicular to the slice thickness. As the beam is rastered, it is continuously or discretely shifted along the direction of the slice thickness with an nominal average linear progression rate in the direction of the slice thickness of vn. After time Δt, a slice of nominal thickness Δln=vnΔt will have been removed. By imaging the fiducial notches at times t0 and t0+Δt , it is possible to determine the actual thickness of the slice Δl or equivalently the actual average progression rate of the beam
and therefore infer the amount of drift of the entire system in the direction of the slice thickness. This drift is primarily comprised of drift of the milling beam due to electronic stability, physical sample drift as well as beam displacement induced by the interaction of the milling and/or imaging beams with the sample.
At certain intervals ti, which might be for every slice of for a number of slices, the effective drift rate of the system vd(ti) can thus be estimated. Given this estimate of the drift rate, it is possible to preemptively and continuously adjust the milling progression rate vm to include the drift rate and therefore produce slices that have thicknesses closer to the target nominal thickness. In the present embodiments, the milling beam progression rate vm is dynamically adjusted at each time ti base of the new measurement of the real progression as well as on past measurements {v(ti)} in an effort to generate the nominal milling rate vn. A reader skilled in the art will recognize that this system is one where current and past measurements are used as feedback to predictively compensate for system errors and instabilities to recover a nominal target.
Such a system can be solved by implementing a control-loop feedback mechanism such as one implemented in a proportional-integral-derivative (PID) controller. A block diagram of a known controller is shown in
The determination of the optimal parameters (KP, KI, KD) for accurately and reliably predicting the proper milling progression rate is beyond the scope of this document. In a simple implementation, the proportional, integral and derivative coefficients of the controller can be fixed by design, and the controller is simply used to calculate the milling rate applied to the beam based on all measurements of slice thicknesses. If the control mechanism is stable, this will result in slice thicknesses that are closer to the nominal target slice thicknesses. This is particularly relevant at very small nominal progression rates when the drift rate is comparatively large. Without compensation, the slice thicknesses could potentially be much too large resulting in loss of sample information.
The position of the fiducial notches can also be used to compensate for system drift as observed by the imaging beam. Prior to acquiring an image, the relative position of the notches is used to determine the z position of the cross-section face, and their absolute position can be used to re-centre the fiducials in the x-y imaging plane. Note that although only a single notch is necessary, using multiple notches allows more flexibility and more robustness in the calculation of the drift. In particular, it can be used to virtually eliminate surface topography effects by choosing which notches to use. It may be beneficial to perform this step during the acquisition process and not simply during post analysis as it allows the target volume of interest to be properly tracked and imaged efficiently. By example, steps 800 and 802 can be executed in combination with step 300 of
Without correction, it may be necessary to image a larger volume than necessary to ensure that the target volume is acquired. Given the potentially very long acquisition times (possibly several days), the system stage may drift several microns under normal operating conditions, so it is advantageous to frequently adjust for this drift to obtain consistent and reliable results.
In addition to statically adjusting the imaging beam shift to re-centre the features prior to acquiring the image, it is also possible to correct for drift during the image based on the current, past and future measurements of the fiducial positions. As a first implementation performed during post-processing, given a measurement before and after an image, a calculation can be performed determine the average drift that occurred during the image and to skew the image to compensate for this drift: if a drift {right arrow over (d)}=(dx, dy) is measured between times t and t+Δt, then an image scanned left to right and top to bottom can be skewed or otherwise adjusted during the scan (or less optimally during post-processing) according to
In an example embodiment, a predictive dynamic drift correction is used to “de-skew” this distortion during scanning. At the end of the scan the predicted drift and the actual drift may be compared, and if necessary, a correction skew may be applied in post processing the image.
The predictive dynamic drift correction uses a Predictive Model of Drift that allows continuous sub-pixel compensation. In implementation, one or more fiducial(s) 1400 are monitored for the time taken to acquire the image 1402. A drift factor is calculated for example, based on a drift of the fiducial(s) 1410) and the drift correction factor is applied on the subsequent frames, for example, to
The dynamic drift correction results in the corrected image scanning the exact desired region, similar if not identical to
In other example embodiments, the dynamic drift correction may be applied based on environmental monitoring. For example, drift correction may be applied as a function of spatial and environmental measurements, such as temperature, pressure, sound etc. The impact of these environmental factors on the drift may be modeled as a function of change in the environmental factor (for example, how a 1 degree change in temperature affects the drift) and the drift correction may be adapted to incorporate these measurements and modeled impact in real time. This enhances the dynamic drift correction capabilities and improves the quality of imaging when an image may take several minutes to acquire.
In a further example embodiment, the dynamic drift correction may also be used to interrupt an image periodically to drift correct. However, the interruption may result in “jumps” in the image and hence it may be preferable to correct the drift as a smooth function based on the predictive model, and perhaps update the predictive model periodically (e.g. multiple times within a scan) instead.
According to a present embodiment, a drift correction method is executed during live imaging as in the case of the preemptive milling correction. Given past drift measurements, a preemptive beam shift is applied during the imaging to compensate for the expected drift during the image, negating the need for post-processing. Again, as in the case of the milling correction, a PID controller can be used to estimate the amount of drift necessary to compensate for the system drift and eliminate any actual shift in the image. For very long image acquisitions, the amount of drift at various times during the image can be re-evaluated (e.g. pause at the end of a line, image the notches, then resume the imaging) in case the required drift correction changed during the acquisition time. Note that the drift correction to be applied can be calculated to a level that is well below the least significant bit on the DAC. For example, the total drift correction calculated to be applied during an imaging sequence can be divided by the toal number of dwell periods during said imaging sequence, and this differential drift correction per pixel can be added to the engine at each and every pixel dwell, and when the cumulative correction becomes large enough to exceed one DAC LSB, the correction effectively shifts the beam by one LSB, and the accumulation of correction continues.
In the examples discussed thus far, the “slice” of material removed by an ion beam is generally removed in a geometry such that the ion beam is at a glancing angle to the surface being sliced. It is also possible to remove material using an ion beam whose angle of incidence is much closer to the normal to the surface being sliced. In general, it is well known in the art (especially from the field of SIMS) that this approach will develop topography, especially if the sample is not homogenous. Gas based chemistries can be used during milling (also known as NanoPatterning or patterning) to enhance the removal rate of a material, sometimes selectively compared to the removal rate of another material, but when the patterning beam deviates from near glancing incidence topographic artifacts as well as artifacts due to uneven gas flow tend to disrupt the uniform removal of material and disrupt the desired “flat bottom” structure necessary for such processes as circuit editing or 3D reconstruction.
The following probabilistic patterning embodiment is used for maintaining a “flat bottomed” structure desired for circuit editing or 3D reconstruction. Most patterning involves repeating the same simple pattern over and over for a given amount of time. In the case of regular shapes such as rectangles, this results in a shape that has been exposed to a uniform dose, and will consequently have eroded or deposited a uniform amount of material, subject to milling or deposition effects (particularly on the periphery of the NanoPatterned areas) that are well known in the art to occur.
To expose a shape to a non-uniform dose, commonly known as “bitmap” or “greyscale” milling, several implementations exist. The most common approach is to vary the dwell time at each pixel of the shape based on the grey level of its corresponding pixel in the shape mask. This results in a dose distribution that is a copy of the dwell time distribution. Another approach is to pattern the shape as a sequence of slices corresponding to the various grey levels in the mask bitmap. This also results in a final dose distribution that matches the grey-level distribution in the mask image. It should be noted that as early as 1995, Micrion Corporation of Peabody, Massachuseets incorporated a means to pattern using this second approach, with the “bitmap” generated at predetermined intervals from the signals (image) generated by the ion beam.
The advantage of the first method is that each pass delivers a proper distributed dose and that all the pixels are visited for each pass, so if the pattern is stopped at any point in time, the actual dose distribution is proportional to the target distribution. However, during each pass, the local dose delivered per pixel is not constant: pixels with longer dwell times receive more dose that those with shorter dwell times. In the case of gas assisted etching or deposition, where the dose per pixel must be closely regulated to avoid gas depletion and optimize milling efficiency, this results in inefficient and sometimes improper milling conditions. The second method does not suffer from this problem since each slice is patterned with the proper dwell and pixel spacings, thereby ensuring gas chemistry to be optimal. However, since the patterning of each slice is occurring as subsequent steps, the proper dose distribution is only obtained after all the slices are processed.
To resolve issues with the existing approaches, the presently described probabilistic patterning method is employed, an embodiment of which is shown in
The result of this approach is that for each pass, the proper dwell time is applied to all of the dwell points, resulting in proper gas assisted processes, and at any time the average local dose will be correct. Based on the central limit theorem of probability theory, the patterned dose distribution at each dwell point will converge to the target distribution given a large enough number of passes. In the present embodiment, the operator has the option to control the number of passes of the pattern by the beam. The greater the number of passes, the closer the actual dose distribution will reflect the target dose distribution.
Although this approach is probabilistic and may be implemented using a number of well known methods of generating random numbers, implementation is accomplished by using a known random number sequence to accept or reject dwell points at each pass, which results in a very deterministic stream of dwell points that can readily be deconvoluted during imaging to determine where the beam dwelled and where it did not, thus allowing the resultant signals from the target to be imaged, thereby allowing the patterning to be constantly monitored in real-time, allowing proper visual endpointing. Indeed, it is possible to apply the approach of Micrion Corporation to determine, via this signal collected at known increments, how to evolve the milled (or otherwise patterned area) over time, based on the image, reconstructed periodically, from the signals and deterministic stream of dwell points.
In addition to resolving both issues of instantaneous dose distribution proportional to target dose distribution, and proper pixel dose per pass, this technique also resolves other issues present when using gas assisted etching. One common example occurs when using XeF2 to assist the etching of silicon, once exposed to the beam, the silicon continues to be etched spontaneously for a short period of time. This spontaneous etching typically is not uniform and results in pitting of the surface which may get accentuated by further milling and will result in a non-flat mill. When using a variable dwell method, the total frame time may be quite long because of the dwell multiplier (a 1000×1000 pixel pattern with a base dwell of 100 ns and an average multiplier of 128 will have a frame time of more than 12 s). When using the slice based approach, in portions of the pattern that do not get many passes, the delay between the slices where the beam is unblanked may be too long, thereby resulting in pitting in these areas.
When using the probabilistic patterning embodiments, even if an area has a low overall probability, the probability that at least one pixel in the vicinity will be patterned at each frame can be quite high. So even though for a given pixel, the time between visits may be large enough to cause problem with self etching if its milling probability is low, the time between visits in its vicinity will be such that self etching can be inhibited. For example, if an area of the pattern has a probability of 20% of being patterned (⅕ of the total dose is desired), then for a square dwell point arrangement where each pixel has 8 neighbours, then the probability of at least one of the neighbours of being visited at each pass is almost 90% (P=1−(1−20/100)9), which direct experimentation has proven to be sufficient to reduce the spontaneous etch issues using XeF2.
A benefit of the probabilistic patterning embodiment is the reduction of the current density of the incident beam on the sample, which can reduce charging artifacts during imaging.
Probabilistic milling can also be used for top-down or nearer normal incidence removal of material for 3D activities, rather than near glancing angle. Scanning the beam in a probabilistic pattern according to the previously described embodiment, rather than in the conventional sequential method, reduces differential milling artifacts that arise from sample features, leading to a more planar removal rate. Note that the sample can be monitored using either the signal generated by the patterning beam during probabilistic patterning, or by another imaging beam such as an electron beam directed at the area of interest. By having the imaging beam off-axis from the patterning beam, a three-dimensional view can be reconstructed whose information can be used to alter the probabilistic patterning probability distribution to smooth out (or enhance if desired) variations arising from sample inhomogeneity.
As previously mentioned, the 2D and 3D imaging embodiments can benefit from additional improvements over control of the CPB system 10, which can be provided by CPB workstation 100. These are referred to as multi-pass rastering, spatial super-sampling and temporal sub-sampling, which can be optionally enabled during the imaging phase in order to improve data quality or optimize a particular component of the signal that is used to generate the image.
During any beam raster operation executed by CPB system 10, which includes imaging, milling, gas assisted etching or deposition, the FIB beam deflection software and hardware deflects or positions the beam in a preset pattern across the surface, generally referred to as rastering. At each preset location, the beam is left to dwell for a given period of time before moving to the next point in the raster. At its simplest, a raster pass consists of positioning the beam at fixed increments along one axis from a start point to an end point, dwelling for a fixed dwell time at each point. At the end of a line, the beam waits a fixed retrace time before moving an increment in a second axis. The beam may return to the start point in the first axis and begin again, or may begin “counting down” the first axis from the point it had just reached (depending on whether the raster type is raster (the former) or serpentine (the latter). This process continues until all increments in both axes have occurred, and the beam has dwelled at all points in the scan. The typical spacing between each point along a raster is determined based on the scan size and the digital scan generator. These factors affect the resolution of the scans as discussed below.
Many CPB systems use 12 bit high speed deflection on a 12 bit scan generator. Dwell time per point is typically less than 1 μs. One example is the Vectra FIB system from FEI Company of Hillsboro Oreg., which is capable of achieving a focused spot with significant beam current that is less than 20 nm in diameter. When operating with a 20 nm spot at a 320 μm FOV at the maximum limit of the 12 bits of the scan generator, the spacing between scan points, ΔxScan and ΔyScan will be about four times the spot size (320 μm/4096≈80 nm). This results in a situation such as that shown diagrammatically in
As can be seen from
The raster scanning method according to a present embodiment improves the spatial resolution of the scans by controlling the beam along a raster. The method advantageously uses the finer placement controls for the beam available on CPB systems. For example, the Vectra can position the beam with much finer placement through a control known as the beam Pan (analogous to a beam Shift or other offset voltage as applied in other CPB systems). On the Vectra and similar systems, the magnitude of the Pan deflection is independent of the field of view, and the minimum Pan increment is on the order of the smallest spot size achievable, although the speed with which the Pan can be varied is typically much slower than the deflection. Thus, even at the 320 μm FOV considered in the earlier example, it is possible to deflect the beam, using the Pan, by an increment that is much smaller than the scan increments ΔxScan and ΔyScan, which are both ˜80 nm at a 320 μm FOV.
Generally, the method may be implemented in an example embodiment as illustrated in the flowchart of
For improving the spatial resolution, one or more offset raster passes are performed 1016. In an offset raster pass, the beam is repositioned the beam at a position offset from the start dwell point in the raster 1016, once the end of the raster is reached. The offset is less than the fixed spacing, and may be determined, for example, as function of the beam size. The beam is then allowed to dwell at the position offset from the start dwell point for the selected period of time 1018. The beam is then iteratively repositioned at subsequent offset dwell points along the raster 1020 and allowed to dwell at each subsequent offset dwell point for the selected period of time 1024 until the end dwell point in the raster is reached 1026. Each subsequent offset dwell point is defined by the fixed spacing from its previous offset dwell point.
At the end of an offset raster pass, if additional offset raster passes are desired 1028, further offset raster passes are performed 1030 and additional offset is applied to the beam 1032 for setting up the next offset raster pass. Upon completion of the multiple-raster passes and no further offsets remain in the current raster, the raster is advanced to the next raster in the scan 1036. Upon completion of multiple-raster passes for each raster in the scan, i.e., when the end of scan is reached 1034, the process is stopped 1040.
With reference to
According to an embodiment of the present invention, this general approach can be repeated a total of 16 times, over raster passes “A” through “P”, sequentially changing ΔxPan and ΔyPan at the completion of each raster. Within each ΔXScan and ΔyScan, sixteen different ΔxPan and ΔyPan could be set, according to Table 1 below:
This would result in a more optimal mapping of the field of view, where after this 16 raster pass operation was complete, the dwell points data could be reconstructed to produce an image where all dwell points were contiguous.
According to another embodiment of the present invention, the previously described method could be further refined by modifying the Pan variation algorithm to operate in a “serpentine” manner, where between each sequential raster pass “A” through “P” only a very small change in the Pan settings would be required. Such small changes could likely be accommodated in a more stable manner by the slower speed Pan electronics. Table 2 illustrates a raster pass mapping according the presently described embodiment of the invention.
This “serpentine” mapping technique preferably uses software configured to activate the necessary ΔxPan and ΔyPan settings between raster passes, as well as to reconstruct the resultant new image at high resolution as a mosaic of 16 separate raster passes. Such software or firmware can be written, or existing control software can be modified.
It is noted that in the above example embodiment, upon completion of the first raster pass (with no offset), the beam is repositioned at the start dwell point with the offset applied. In other example embodiments, the beam need not be repositioned at the start dwell point (with the offset) for the subsequent raster, but the subsequent raster may commence at the end dwell point, by positioning the beam at a position offset from the end dwell point and rastering in a direction opposite to the previous raster. This process is referred to as “double serpentining.”
An advantage of this technique is the virtue of requiring no major changes to the deflection electronics. It is noted that if the deflection power supplies prove to be insufficiently stable, they can be replaced with more stable units without effecting the other components or control logic. It is further noted that the presently described embodiments are not limited to 16 raster passes.
Another advantage of this embodiment is the fact that, as the CPB system is designed to accept Pan changes during rastering, this method could be implemented without requiring a change to the raster control software merely by setting up a system to set the necessary ΔxPan and ΔyPan settings at the appropriate points in time. Otherwise, the raster control software would operate normally and yield the correct values for dose per unit area, etc. As CPB systems such as the Vectra are designed with a “refresh” interval, whereby the beam pauses for a definable amount of time at the end of each raster pass, setting the appropriate ΔxPan and ΔyPan settings could be accomplished in software during the refresh time at the end of a raster. Another approach would be to examine the vertical retrace signal generated in hardware by the raster generator and make appropriate modifications to the Pan values when a vertical retrace signal was detected.
Dwell point analysis software such as FIB Assist from Fibics Incorporated of Ottawa, Canada could be configured to assemble the appropriate “human-readable” high resolution images from such an approach as well as set the necessary ΔxPan and ΔyPan settings at the appropriate points in time. By providing an appropriate user interface, such a system could theoretically achieve 20 nm spatial and placement resolution anywhere within a 320 μm (+/−160 μm) field of view without resorting to stage motion, on the existing Vectra system electronics.
An example implementation of the aforementioned embodiments of the present invention is now be described.
In this method, all image and mill commands generated by the user are “filtered” by control software, which enable the user to position the stage at a fixed point and operate within a 320 μm FOV, moving as if they were moving the stage, but without stage motion. A typical implementation entails the user performing imaging and enabling mills anywhere within that 320 μm FOV, at “Effective” Fields of View (EFOV) from less than 1 μm up to 320 μm in this mode.
A user request for a standard image pass for a 1,024×1,024 image is intercepted by the CPB workstation 100 and turned into a request for 16 rasters of a mill with 256×256 dwell points, plus implementation of the necessary Pan adjustments (4 adjustments to the ΔxPan and 4 adjustments to the ΔyPan) in an automated fashion. Note that the number of Pan adjustments required for these algorithms are small compared to the full Pan range, so the user would not see a significant reduction in the Pan range available to them through implementation of this technique. The process of defining and building the image is handled in software and is virtually transparent to the user.
To simplify visualization of such an approach, consider an EFOV of 10 μm that can be “scrolled” to anywhere within +/−160 μm of the stage center in High Resolution Navigation (HRN) mode. Using appropriate controls, the user enters this HRN mode, and stage readback, Knights navigation, EFOV, etc. would function as if the user was moving the stage.
The system can optionally apply the Pan corrections on a line by line basis during imaging, in order to generate full resolution lines one at a time, rather than, as in the scheme, above full resolution frames one at a time. In other words, it may be preferable to perform multiple repeats of the same line with ΔxPan corrections applied during the horizontal retrace until the full resolution line is composed and displayed to the user, before proceeding to the next line.
The Vectra possesses the necessary “Line Scan” algorithm to raster a mill while delivering the full dose to each line before proceeding to the next line. In this case the user sees a high resolution image built up on a line by line basis, that appears identical to the image that is formed during a conventional image pass. This image is responsive to focus and stigmation in the same manner as a conventional image. Such a line by line process can be applied to milling as well as imaging, however the scheme outlined above may be more suitable to the very short dwell times required for gas assisted milling operations, whereas the longer dwell times typical of imaging operations could more easily support the formation of the high resolution image on a line by line basis.
By appropriately intercepting all calls for imaging and milling operations at a given EFOV, and recasting them to incorporate the Pan adjustment scheme and data processing at the real FOV, the user will not be aware of the “machinations” occurring in the background, but would instead appear to have gained a 16 bit (or higher) deflection system where they previously had only 12 bits. On systems with a higher native DAC resolution, this approach can yield similar improvements.
To simplify the process and avoid errors arising from the granularity of the Pan deflections themselves, the number of EFOVs available in HRN mode can be fixed to a few EFOVs at optimal values ranging from 0.25 μm, to 320 μm, for example. Note that using this approach it will be possible to obtain EFOVs less than 1 μm on the Vectra, however whether in the conventional “real FOV” approach or the modified “EFOV” approach described above, the information limit of all FOVs is ultimately determined by the spot size of the microscope (given sufficient stability of all other components and a suitable specimen).
It should be noted that the “Pan” described above need not be considered as solely the use of the “Pan” or “beam Shift” of the microscope, but could instead be a further offset applied in digital or analog space within the DAC subsystem, or a raster subsystem based on one or more DACs. The described method of rastering the beam across the field of view in a number of discrete steps across multiple passes, ie. dwells at cyclical points A through P, as shown in
Thus, this method can be an effective method to reduce artifacts such as sample charging, drift, contamination and beam damage. One skilled in the art will realize that the actual granularity need not be 4×4 dwell points (i.e. sixteen passes is not a “magic” number), and that alignment techniques such as are commonly employed for “drift correction” in CPB systems may be necessary to align the “center of mass” of each image pass to improve the overall result in the face of whatever “drift” may occur in the imaged area over the time it takes to acquire these multiple passes. Indeed, the total time to acquire 16 passes each at 1/16th of the number of dwell points will be very similar to the time to acquire a single image pass in the standard manner, however in the standard manner the drift will be distributed throughout the image (resulting in a potential “stretching” of the image features) whereas in the method described here, for an equivalent amount of drift per unit time, this stretching will be smaller (on the order of 1/16th of the stretch) per pass, and with application of drift correction between passes to realign on the field of view of interest, there can be a significant improvement in fidelity, as well as improvement due to the reduction in charging. It should be noted that the derived granularity of sub-positioning the beam may be in a “regular” fashion as described or may be accomplished using probabilistic methods similar to those described above.
Another method to improve the scan quality when acquiring images with pixel spacings much larger than the spot size of the beam is a spatial super sampling method, illustrated in the flowchart of
The presently described method advantageously uses the pixel's intensity as a function of the average intensity of the area represents by the pixel in the image, rather than a single sampling of the area covered by the beam itself. In order to generate this average intensity, the beam can, during the dwell time of that pixel, be moved around randomly or systematically within the pixel sub-area. This is known as spatial super-sampling.
Although from an imaging point of view this may be partially accomplished by defocusing the beam to match the pixel size, defocusing may not be suitable when patterning with gases as it affects the spot current density. In the case of gas assisted etching or deposition, very large pixel spacings are commonly used to improve the gas efficiency, but this leads to non-uniform milling or deposition. By moving the beam around with sub-pixel resolution during the patterning, a more uniform etch or deposition can be obtained without sacrificing the efficiency of the gas process. It should be noted that such movement may be in a regular fashion or using probabilistic methods similar to those described above.
As illustrated in
In an example embodiment, the spatial super-sampling may be implemented in a digital scan-generator by over-clocking the output DAC at a rate significantly higher than the dwell time. The nominal scan data can then be shifted by a random or fixed amount along either scan axis. The shifted beam position is clocked out several times around each nominal pixel position. For example, by updating clocking the DAC at 50 MHz (new data every 20 ns), it is possible to generate 50 distinct sampling locations during a 1 μs pixel dwell time, thereby spatially super-sampling the dwell area within the dwell time.
It is noted that the benefit is not limited to a case where the pixel spacing is much larger than the spot size. The benefit may also be realized by applying this technique under other conditions where the spot size is near or larger than the pixel size.
Another method to improve the scan quality when acquiring images is by temporal sub-sampling. According to this embodiment, scan quality is improved by extracting the signal variation in time once a dwell point is irradiated by the beam. The intensity of a pixel is normally obtained by summing and averaging the detector signal during the entire time the beam is dwelling at that location. In cases where the beam shortly interacts with the sample, it may be useful to extract how the signal varies in time once the sample dwell point is irradiated by the beam. This data can be used to extract dynamic process information, or to exclude one or more time slices during which there is an extraneous or otherwise undesired signal.
In an example embodiment, this can be implemented by sampling the intensity data at a higher rate than the dwell time in order to get access and process this data. For example, the system may sample the intensity at a frequency of 40 MHz, which produces a sample every 25 ns. Under normal circumstances, these samples are accumulated during the entire dwell period to generate an average intensity. When the information of interest is only in the signal after the first 200 ns, the first 8 samples might be rejected, and all subsequent samples can be averaged, integrated, or otherwise processed to produce the displayed intensity.
By way of another example, the initial signal from an ion beam may in fact contain information on the chemical state at the surface that is lost after the first few ions have impacted. It may be advantageous to combine both methods, by splitting the initial and subsequent data, or any number of time slices within the dwell period. Although processing the intensity data in this fashion can be done in a purely analog system, it is easier to implement a flexible solution in a digital system where the intensity is sampled by an ADC at high speed and processed by an FPGA or DSP prior to being displayed. This could also be accomplished in software on a computer if the entire high speed data stream is collected and processed prior to being displayed.
The signal may be separated into different components for identifying properties of the sample—for example, chemical state, charge state, capacitive contrast effects, etc. based on time slices, or on the variation of the signal within the spatial super-sampling. Also, the entire super-sampled data may be set and subdivide based on characteristics (rise, fall, slope, noise level, etc.) rather than purely on time slices.
The previously described embodiments for maneuvering the beam involves deflecting the beam a given amount in X and Y axes so that the beam strikes the target at a nominally known position. One method of accomplishing this involves applying a voltage to a series of plates or coils to deflect the beam in X and Y axes, with the magnitude of the voltage correlating to the magnitude of the deflection. This particular aspect of CPB systems should be well known to those skilled in the art. Historically, this deflection was produced by analog circuitry, and this is still the case in many systems on the market today.
More recently, systems have been marketed where the deflection position was determined using a digital scan generator, and a digital to analog converter (DAC) was used to produce the deflection voltages in response to a digital deflection code.
Many of the initial digital deflection systems and their DACs were based on 12 bits in both X and Y axes, yielding 4,096 discrete positions that the beam could be deflected to, assuming sufficiently fast and stable electronics and power supplies. These 12 bit DACs typically had the virtue the difference in the analog output values resulting from a unity change in the digital code applied to the DAC, say from code N to code N+1 (effectively a change of one Least Significant Bit (LSB)) deviates from the ideal difference of the analog output by no more than the ideal difference itself. Mathematically, DNL=Max(|(Vout(i+1)−Vout(i))−Videal LSB step|), and these 12 bit DACs typically were specified such that the DNL error was less than or equal to 1 LSB, thereby guaranteeing a monotonic transfer function with no missing codes.
When observing a 20 μm field of view in such a FIB system, 12 bits were sufficient, as the 20 μm field of view (FOV) would be broken down into 4,096 discrete positions, effectively mapping each position with a square just under 0.005×0.005 μm (5 nm×5 nm) in size. As the best beam resolution achievable was on the order of 5 nm, this degree of granularity was sufficient for a 20 μm or smaller FOV.
When high placement resolution was required at sites outside the 20 μm FOV, it became necessary to physically move the stage in order to reposition the new site(s) within the 20 μm FOV, or to use a larger field of view and accept the poorer placement resolution available. For example, to work on two sites 200 μm apart, one can either
(a) shuttle between the two sites with stage motion and continue working with a 20 μm field of view, 5 nm placement accuracy, and any stage positioning error that may occur, or
(b) increase the field of view to 200 μm, removing the need to move the stage and introduce a potential inaccuracy in stage motion, but instead the user must accept a 10 times poorer placement accuracy of just under 50 nm.
Newer systems typically employ DACs and scan generators based on 16 or more bits, to allow greater placement accuracy at larger fields of view. Another approach that will work with a 12 bit ADC is to define a fixed offset voltage that deflects the center of the field of view a known amount, and shuttle between points using this offset voltage rather than stage motion, while retaining the 5 nm placement resolution.
In the CPB systems, including FIB and SEM systems, a digital to analog converter (DAC) is used to convert a code into a corresponding voltage magnitude for application to the system deflection plates. Given that the beam can be deflected in the X and Y axes, separate X and Y deflection codes are provided by the control system. There is a range of available codes spanning a min code value and a max code value, where each code is calibrated to provide a predetermined deflection voltage. In some systems, the scan generator and the DAC's are configured based on 16 bit codes. Ideally, the deflection voltages from the min code to the max code follow a linear relationship. In the presently described CPB system, either a single DAC is used to generate both the X and Y deflection voltages, or dedicated DAC's are used for generating the X and Y deflection voltages. In some CPB systems, the scan generator and the DAC's are mounted to a daughterboard, which in turn is connected to a motherboard of the system.
Ideally, to achieve a monotonic transfer function with no missing codes at the 16 bit level, the differential non linearity (DNL) of the DAC and raster generator subsystem should not exceed 1. While the DAC may have a native DNL, additional circuitry on the daughterboard can increase the total DNL. It should be understood that a lower DNL is desired. DAC systems at greater than 16 bits do exist that guarantee a DNL of less than 1 LSB at the 16 bit level, however there is another requirement for this application—the DAC preferably outputs at a minimum frequency on the order of 40 MHz (25 ns dwell times). No “high speed” DAC commercially available at this point has a DNL of less than 1 across all digital codes and can also operate at these speeds. Some high speed DAC integrated circuits do come close, and can have average DNL value which is less than one LSB with very low standard deviation, however experimental testing has discovered that large variations in the DNL often occurs at the code boundaries which are certain powers of two, which is likely caused by the DAC architecture. By example, the most pronounced DNL variations observed in some DAC's occur at code boundaries that are multiples of 4,096. In otherwords, certain input codes for the DAC will generate a voltage that is non-linear with the voltages generated by the other codes. This is not unexpected given the architecture of certain DACs which are comprised of “strings” of resistors, each responsible for a portion of the full slope of the output analog value; at the points where these resistor strings must be matched, it is more difficult to achieve a low DNL. It should be clear that such variations will contribute to scan inaccuracies by the CPB system. For example, in the system of
DNL measurements for every input code of a commercially available DAC device are shown in the graph of
The effect of such abnormal DNL spikes can be mitigated by taking advantage of the fact that output voltages corresponding to codes proximate to a code having an abnormally high DNL value, will typically have low DNL values. According to the present embodiments, the output voltage corresponding to the target code of interest having an abnormally high DNL value is averaged with the output voltages corresponding to codes proximate to the target code.
Multi-DAC voltage generator 1300 includes three identical n-bit DAC devices 1302, 1304 and 1306, and a voltage averager 1308. DAC device 1304 receives the target Y input code Y_CODE generated by the raster generator. DAC device 1302 receives a Y input code that is one code step above the target Y input code, and is referred to as Y_CODE+1. A code step is the subsequent code to a target code or the preceding code to a target code. DAC device 1306 receives a Y input code that is one code step below the target Y input code, and is referred to as Y_CODE−1. Both Y_CODE+1 and Y_CODE−1 can be generated automatically by the control system in response to Y_CODE simply by incrementing Y_CODE by one code step and decrementing Y_CODE by one code step. Accordingly, DAC 1304 generates a voltage Vy, DAC 1302 generates a voltage Vy+1 and DAC 1306 generates a voltage Vy−1. Voltage averager 1308 receives all three output voltages and provides an output voltage Vy_AVG representing the average of voltages Vy, Vy+1 and Vy−1. Therefore, all three DAC devices operate in parallel, but with different input codes.
According to the principles of the presently shown embodiment, if the target code Y_CODE happens to have an abnormally high DNL, then the output voltage for DAC 1304 is averaged with the voltages provided from the other two DAC devices having input codes adjacent to the target code. Because the DNL for the other input codes adjacent to the target codes have normal/low DNL, the output voltages from DAC devices 1302 and 1306 will have normal voltage levels expected for those codes. Thus the resulting Vy_AVG voltage for the corresponding target code becomes closer to the expected level. As previously mentioned, an identical circuit can be used for generating the X deflection voltage.
The circuit embodiment of voltage averager 1308 shown in
In order to illustrate the effectiveness of the presently shown embodiments for all target codes, DNL measurements for every input code for the Multi-DAC voltage generator 1300 are shown in the graph of
It is noted that the number of DAC's used in the embodiment of
According to further alternate embodiments, the codes received by the Multi-DAC voltage generator 1300, or its previously described alternate embodiments, are not limited to receiving codes that are one code apart from each other. More specifically, the DAC's can receive codes that are two or more codes apart from each other. For example in the embodiment of
In the preceding description, for purposes of explanation, numerous details are set forth in order to provide a thorough understanding of the embodiments. However, it will be apparent to one skilled in the art that these specific details are not required. In other instances, well-known electrical structures and circuits are shown in block diagram form in order not to obscure the understanding. For example, specific details are not provided as to whether the embodiments described herein are implemented as a software routine, hardware circuit, firmware, or a combination thereof.
Embodiments of the disclosure can be represented as a computer program product stored in a machine-readable medium (also referred to as a computer-readable medium, a processor-readable medium, or a computer usable medium having a computer-readable program code embodied therein). The machine-readable medium can be any suitable tangible, non-transitory medium, including magnetic, optical, or electrical storage medium including a diskette, compact disk read only memory (CD-ROM), memory device (volatile or non-volatile), or similar storage mechanism. The machine-readable medium can contain various sets of instructions, code sequences, configuration information, or other data, which, when executed, cause a processor to perform steps in a method according to an embodiment of the disclosure. Those of ordinary skill in the art will appreciate that other instructions and operations necessary to implement the described implementations can also be stored on the machine-readable medium. The instructions stored on the machine-readable medium can be executed by a processor or other suitable processing device, and can interface with circuitry to perform the described tasks.
The above-described embodiments are intended to be examples only. Alterations, modifications and variations can be effected to the particular embodiments by those of skill in the art without departing from the scope, which is defined solely by the claims appended hereto.
This application claims the benefit of priority of U.S. Provisional Patent Application No. 61/485,713 filed May 13, 2011, which is incorporated herein by reference in its entirety.
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/CA12/50316 | 5/14/2012 | WO | 00 | 11/12/2013 |
| Number | Date | Country | |
|---|---|---|---|
| 61485713 | May 2011 | US |
