Patent Application
20050229053
1. Field of the Invention
The present invention relates, in general, to testing the parameters of high frequency signal waveforms and, more specifically, to testing the parameters of high frequency signal waveforms using low frequency measurements and circuitry.
2. Description of Related Art
As the data rate of integrated circuit (IC) pins increases each year, to many gigabits per second, it becomes beneficial to develop test methods that do not require test equipment to operate at the pin data rate being tested.
As shown in
Referring to
For differential signals, a termination resistor, comprising resistors 26 and 28, as illustrated in
The power of arbitrary high frequency signals, especially radio frequency (RF) signals, is commonly measured via a diode 34, shown in
Properties that are typically measured for a high frequency data signal waveform shape include logic levels (voltage of logic 1 and logic 0), rise and fall transition times (measured at 10% to 90%, or 20% to 80%, of the interval between the logic 1 and logic 0 voltage levels), overshoot and undershoot (excess or insufficient voltage immediately following a transition), duty cycle distortion (difference between the width of an isolated group of consecutive logic 1 bits and the ideal width), and pre-emphasis (intentional, temporary, excessive signal level changes for every change in logic level).
The present invention seeks to test these and other waveform shape properties of high frequency data signals using only low frequency (LF) test equipment and test access circuitry.
The present invention is used to measure properties of a relatively high frequency data signal waveform. The properties include, but are not limited to, the logic levels of various bit positions in a periodic sequence of bits, rise and fall transition times, some types of overshoot and undershoot, duty cycle distortion, and pre-emphasis level. High frequency data signals refer to signals having data rates in the range of up to many gigabits per second. This is achieved by measuring average voltage and average voltage squared for a waveform based on various data patterns and then performing calculations to deduce the waveform properties.
One aspect of the present invention is generally defined as a method of deducing properties of the shape of a waveform, comprising the steps of (a) generating a signal based on a periodic pattern of logic levels; (b) measuring a DC level that is proportional to the average level of the signal and a DC level that is proportional to the average of the signal level squared; (c) repeating steps (a) and (b) one or more times; and (d) calculating a property value of the shape of the waveform based on a plurality of measurements.
Another aspect of the present invention is generally defined as a circuit for deducing properties of the shape of a waveform comprising: a circuit for generating a signal based on a periodic data waveform; a circuit for generating a DC level proportional to the average of the waveform level; a circuit for generating a DC level proportional to the average of the waveform level squared; a circuit for DC level measurement; a circuit for storing DC measurement values; and a circuit for calculating a property of the waveform's shape based on a plurality of measured DC values.
The method and circuitry can be used for digital signals with two or more logic levels, for voltage, current, optical and other types of signals, for other properties of a waveform shape, and for analog signals that convey digital data.
These and other features of the invention will become more apparent from the following description in which reference is made to the appended drawings in which:
In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, However, it will be understood by those skilled in the art that the present invention may be practiced without these specific details. In other instances, well known methods, procedures, components and circuits have not been described in detail so as not to obscure aspects of the present invention.
As indicated earlier, the present invention seeks to test properties of high frequency signal waveform shapes by only measuring DC voltages. Advantages of testing the properties via DC voltages include: DC voltages can be measured accurately (within tens of microvolts) and quickly (in less than a millisecond) in the presence of substantial noise; DC test access circuitry is simpler to design, has less impact on the signal under test, and is more tolerant of manufacturing process variations than high frequency test access circuitry.
The Circuit
As shown in
Linear Access Circuits
Any of several access circuits can be used to facilitate measurement of the average voltage of an HF signal, and three of these will be described in the next three paragraphs.
In a first linear access circuit 70, shown in
In a second linear access circuit 76, shown in
In a third linear access circuit 82, shown in
Square-Law Access Circuits
Any of several access circuits can be used to permit measurement of the average squared voltage of an HF signal, and three of these are described in the following paragraphs.
In a first square-law access circuit, one of the source or drain of an MOS transistor is connected to the HF signal, the other of the source or drain is connected to a load resistance or to a virtual ground, and the gate of the transistor is connected to a DC voltage. A similar connection arrangement is described in Khoury et al. U.S. Pat. No. 4,835,421 granted on May 30, 1989 for “Squaring circuits in MOS integrated circuit technology”. However, that arrangement is only suitable for differential signals. The current flowing between the source/drain will be a polynomial function of the high frequency signal voltage. Typically, if the MOS transistor channel length is not “deep sub-micron” (i.e., it does not exhibit what are commonly known as short channel effects), the first-order (linear) and second-order (square-law) terms of the polynomial will be most significant. For deep sub-micron transistors, the zeroth order (constant) and third-order terms can also be significant.
In a preferred second square-law access circuit 82, as shown in
In a third square-law access circuit, a diode is connected between the signal and a resistor to ground, as shown in
Additional transistors 100 may be connected in series with any of the above square-law access circuits, as shown in dotted lines in
Referring to
A combination of the above linear and square-law access circuits may be used. For example, when the accessed circuit node has a steady-state DC voltage, the voltage can be measured via a series transistor or resistor to accurately determine its DC value, independent of transistor manufacturing process variations, while simultaneously measuring the output of the transistor gate access circuit to determine the transistor's linear or square-law gain. When the signal at the accessed node becomes a high frequency signal, the output of the series transistor will no longer accurately indicate the linear average due to the transistor's non-linear resistance, but the current through the transistor whose gate terminal is connected to the accessed node will be more accurately proportional to the average voltage of the high frequency signal, and the proportionality constant (the gain) will be known for this particular transistor. The gain for each test access transistor is preferably measured because transistor gain can vary between ICs and within a single IC. The value of each measured DC voltage is stored until sufficient periodic waveforms have been generated by the signal generator, and then calculations are performed to estimate the value of properties of the waveforms.
The Method
Using the circuitry described in the preceding paragraphs, the properties of a high frequency waveform can be measured and tested using DC measurements, according to the method of the present invention.
As previously mentioned, the method of the present invention generally comprises the steps of (a) generating a signal based on a periodic pattern of logic levels; (b) measuring a DC level that is proportional to the average level of the signal and a DC level that is proportional to the average of the signal level squared; (c) repeating steps (a) and (b) one or more times; and (d) calculating a property value of the shape of the waveform based on a plurality of measurements.
As explained more fully later, the logic voltages for the Mth bit position in a series of M (or more) consecutive bits is deduced by measuring the average voltage for a periodic pattern containing M consecutive bits of the same logic value, then measuring the average voltage for a periodic pattern that is the same except that it contains M-1 consecutive bits of the logic value, and then performing a calculation using the two measured voltages.
Overshoot or undershoot for a rising transition is deduced by measuring the average voltage for a periodic pattern containing consecutive logic 0 bits split into two groups separated by a single logic 1 bit, and comparing the calculated logic voltage value for the single logic 1 bit to the previously deduced logic voltage values for the Mth logic 1 bit, where M>1, in a sequence of consecutive logic 1 bits. An analogous measurement can be done for a single logic 0 bit.
The sum of the signal rise and fall times is deduced by measuring the average squared voltage for a periodic pattern containing M consecutive bits of the same logic value, then measuring the average squared voltage for a periodic pattern containing the M consecutive bits split into two groups of consecutive bits, and then performing a calculation using the two measured voltages. The difference between the signal rise and fall times can be deduced, for some waveforms, by measuring the linear average for the same two waveforms, and then performing a calculation using the two measured voltages. For other waveforms, the difference cannot be calculated from these two measurements, but the duty cycle distortion can.
The amount of pre-emphasis is deduced by first deducing the sum of the rise and fall time for the signal without pre-emphasis, and then measuring the average squared voltage for the same two periodic patterns with pre-emphasis applied, and then performing a calculation using the two measured voltages and the deduced rise and fall times.
The following paragraphs describe example procedures of the general method, shown in
Referring to
In accordance with the method of the present invention, a first signal is generated based on a first periodic digital pattern (step 124) and the average of the signal level and signal level squared are measured (step 126). Then, a second signal is generated based on a second periodic digital pattern (step 128) and the average of the signal level and signal level squared of the second signal are measured. (step 130). Then, based on the measurements obtained in steps 126 and 130, a property of signal waveform shape is calculated (step 132). Steps 124 to 132 may be repeated as needed for other properties.
The method may be better understood from the examples described below. It will be understood from the examples that some properties can be measured without involving the square-law measurements, although most properties require both the linear and square-law measurements.
Before understanding how the properties can be measured, it should be understood that, generally, the properties of a high frequency waveform can only be estimates. For example, the logic 1 level for a digital signal depends on where in the waveform the property is measured. As shown in
Logic Level Voltage
To deduce the logic levels 146, 140, 148, 150 of a signal, a periodic data pattern is first generated containing a sequence of consecutive logic 1 bits, for example 1111000100, as shown in waveform 160, V1, of
Next, a periodic data pattern is generated containing a sequence of consecutive logic 1 bits, where the number of consecutive logic 1 bits is different, for example one more logic 1 bit, as shown in the 1111100100 waveform 162, (V2), of
The voltage difference between the logic 1 voltage and the logic 0 voltage is estimated as follows:
Vlogic1−Vlogic0=N×(V2avg−V1avg)/(M2−M1), where
The values of Vlogic0 and Vlogic1 are estimated as follows:
Vlogic0=V2avg−(Vlogic1−Vlogic0)×M2/N
Vlogic1=Vlogic0+(Vlogic1−Vlogic0)
If the measurements are for each signal of a differential pair, then the resulting voltage estimates can be subtracted from each other to estimate the differential voltage:
Vlogic1−Vlogic0)differential=(Vlogic1−Vlogic0)non-inv−(Vlogic1−Vlogic0)inv
The procedure can be performed repeatedly, each time adding (or removing) a logic 1 bit (or bits) from the sequence of logic 1 bits. This permits estimation of the logic level for each bit position in the sequence.
The calculated values will be accurate if rise and fall transition times are less than the duration of the sequence of 1's , and for unequal or equal rise and fall times.
The bit pattern that can be transmitted is typically programmed, although sometimes an encoding circuit exists and must be disabled for this test. For example, the standard 8B/10B coding scheme converts eight bit data words into ten bit words in which the number of consecutive same-value bits is limited to five and the total number of logic 1 bits in pairs of ten bit words is maintained at ten to minimize the variation in the average voltage of the signal.
Duty Cycle Distortion
To deduce duty cycle distortion, DCD, a periodic data pattern is first generated containing a single, maximal length sequence of consecutive logic 1 bits separated by a maximal length sequence of consecutive logic 0 bits. For example, when N=10, the periodic data pattern could comprise 1111100000, as shown in waveform 164 (V3) of
The duty cycle distortion for logic 1 bits relative to logic 0 bits, is estimated as follows:
DCD1=N×(V4avg−V3avg)/(Vlogic1−Vlogic0), where
If the DCD is known to be insignificant by design (less than 0.01 UI), and the rise and fall times are dominated by the rise and fall time of the output stage of the signal generator, then the difference in the two average voltages is due to the difference between shaded triangles 168, 170 of waveform 166 V4, and the difference between the rise and fall times can be estimated (using the same pair of periodic patterns and measured average voltages) with the following calculation:
tRISE−tFALL=P×2N×(V3avg−V4avg)/(Vlogic1−Vlogic0), where
When the rise and fall times are equal, the shaded triangles of waveform 138 V4 are equal (except the rise triangle area 168 is excluded, hence negative, and the fall triangle area 170 is included, hence positive), and hence the sum of their areas is zero regardless of the transition time.
Transition Times
A more general method for measuring transition time involves measuring the average voltage squared. The shaded portions for the rise and fall transitions of the squared waveform 172 V4sq are not equal when the rise and fall times are equal, and hence the average voltage can reveal their difference. The same patterns can be used from the previous example, and their squared voltages might look like those of waveforms 174 V3sq and 172 V4sq in
(tFALL+tRISE)/2=P×B×N×(I3avgsq−I4avgsq)/(4×Icurve), where
If the rise or fall time is greater than 1 UI, and DCD is insignificant, then an alternative waveform and calculation can be used. A periodic data pattern is generated containing isolated logic 1 bits separated by a maximal length sequence of consecutive logic 0 bits. For example, when N=10, the periodic data pattern could comprise 1000010000 as shown in waveform 176 V5 of
tRISE=P×(1±(1−K(tRISE−tFALL))0.5)/K, where
The range in logic 1 values for a five bit sequence of logic 1 bits can be estimated by extrapolation. The range from bit 1 to bit 5 is likely to be twice the range from bit 3 to bit 5. For example if the third bit has Vlogic1=1.1 V, and the fifth bit has Vlogic1=1.0 V, then the estimated Vlogic1 range is 2×(1.1−1.0)=0.2 volt.
The logic 1 value for the first bit of a sequence, or an isolated logic 1 bit, can sometimes be estimated by comparing the estimated Vlogic1 for the second bit in a sequence of logic 1 bits with the estimated Vlogic1 for the third bit (or a subsequent bit). This is because rise transition undershoot or overshoot that is caused by the driver and not by transmission line effects (ringing) can cause the calculated Vlogic1 for the second bit to appear too large or too small respectively, and it is unlikely that the second bit would be significantly different than subsequent bits. The procedure comprises the following: a periodic data pattern is first generated containing a sequence of three consecutive logic 1 bits, and the average voltage V3avg is measured; then a pattern containing a sequence of two consecutive logic 1 bits is generated, and the average voltage V2avg is measured; and then a pattern containing a single logic 1 bit is generated, and the average voltage V1avg is measured. Vlogic1−Vlogic0 is first calculated for bit 3 and for bit 2, using the method described previously:
Vbit3=Vlogic1−Vlogic0=N×(V3avg−V2avg)
Vbit2=Vlogic1−Vlogic0=N×(V2avg−V1avg)
If Vbit2 is significantly greater than or less than Vbit3, the logic swing Vbit1 is estimated as follows:
Vbit1=Vbit3−(Vbit2−Vbit3)
Pre-Emphasis
As shown in the ideal waveform 180 of
A=N×(I4preavgsq−I3preavgsq)/(16×Icurve) where:
Note that the equation for pre-emphasis, A, is similar to the equation for tRISE+tFALL. They differ by a factor of approximately 10 (for a linear ramp transition, measured at 20% and 80% points), and the sign is opposite (because the terms I4avgsq and I3avgsq are interchanged around the minus sign).
When rise and fall times are significant, they are first deduced using the tRISE+tFALL measurement procedure described earlier, and then the estimated value for pre-emphasis is adjusted to account for the rise and fall time. The resulting calculation is:
A=[(I4preavgsq−I3preavgsq)+2(I4avgsq−I3avgsq)]/[(16×Icurve)/N+2(I4avgsq−I3avgsq)(Vlogic1−Vlogic0)], where the terms are as defined previously.
The mathematical equations presented herein are examples of how the method of the present invention can be used to deduce the waveform of a signal via DC measurements of the periodic waveform's linear average and square-law average. Experiments reveal that other mathematical relationships can be derived for various categories of waveforms and assumptions. Characterization of a circuit's waveforms, followed by correlation analysis, can produce other mathematical equations relating the DC measurement values and the waveform's properties. Genetic algorithms are an example of a systematic way to find these mathematical equations for particular circuits being tested.
In summary, the procedures described can be performed in succession and comprise only measurements of DC voltages for different digital patterns. For data rates above 1 Gbit/second, the averaging can be performed with a first-order RC low pass filter that has a time constant of a few microseconds to permit a sufficiently stable average voltage to be measured in less than fifty microseconds. Voltage is only an example of the signal waveform property that can be measured using the present invention—the same general circuit and method can measure a current waveform, an optical signal waveform, a magnetic field waveform, and others, because each of these signal types has square-law circuits in the prior art that have been developed to derive a DC or low frequency level that is proportional to the power of the signal waveform.
These procedures can be used to test these parameters for any circuit that conveys DC levels, including some analog circuits. For a circuit that does not convey DC levels, such as the capacitor-coupling shown in
By dividing the deduced logic levels of a circuit's output by the values deduced for its input, the linear voltage gain of the circuit can be deduced. Any increase in the deduced transition times can be used to calculate the circuit's frequency response, and any decrease in only the deduced transition times can be used to calculate the non-linear voltage gain (linear gain followed by hard limiting).
The method can be applied to the determination of logic voltages for signals that have more than two voltage levels, by changing selected bits and measuring the resultant change in average voltage.
For all of the tests described herein, test limits for the values calculated may be determined by characterizing known good devices and known bad devices. Test limits may be pre-calculated for the last measurement in each procedure so that a circuit under test can be immediately passed or failed after the measurement.
The important capability provided by the circuit and method of the present invention is the ability to quickly and accurately measure at-speed logic levels without needing high frequency access or high frequency measurement capability. Prior art circuits and methods are not able to achieve this accuracy without requiring very accurate passive components and/or very high bandwidth test access.
Although the present invention has been described in detail with regard to preferred embodiments and drawings of the invention, it will be apparent to those skilled in the art that various adaptions, modifications and alterations may be accomplished without departing from the spirit and scope of the present invention. Accordingly, it is to be understood that the accompanying drawings as set forth hereinabove are not intended to limit the breadth of the present invention, which should be inferred only from the following claims and their appropriately construed legal equivalents.
This application claims the benefit of U.S. Provisional Patent Application Ser. No. 60/489,902 filed on Jul. 25, 2003.
| Number | Date | Country | |
|---|---|---|---|
| 60489902 | Jul 2003 | US |
