This disclosure relates to characterization of a material using multiple modalities in one or more environments. This disclosure also relates to predicting one or more properties of materials or environmental conditions based on the characterization of the material using one or more modalities at a set of other environmental conditions.
Materials may be characterized using one or more modalities. For example, the modalities may be electrical, optical, gravimetric and viscoelastic, and mechanical. In a known system, when multiple modalities are to be tested, an operator initially provides acquisition parameters for one of the modalities and subsequently acquires the measurements using the acquisition parameters. Afterwards, the measurements are processed including any calculations of other properties, interpreted, and displayed. This process may be repeated with other acquisition parameters for the same modality. Once one of the modalities characterizations is completed, another modality is separately tested, data acquired and processed.
However, in a typical system, the acquisition parameters are set in advanced and are not adjusted dynamically based on previous measurement responses (or calculated) or environmental conditions. For example, a known quartz crystal micro-balance with Dissipation Mode (QCM-D) uses a single frequency window of all QCM-D frequencies and with a set resolution. However, the use of a single frequency window that is large (with a high resolution) has a long data acquisition time to acquire frequency data over the large single frequency window. Additionally, since there is a frequency shift, and peak broadening at higher overtones (harmonics), there may be a loss of the full peak or an incorrect fitting where a peak is not completely detected. Alternatively, a reduction in the resolution, e.g., number of frequencies acquired within the single frequency window, leads to a loss of sensitivity.
Moreover, under certain environmental conditions, such as relative humidity, there may significantly decrease the intensity and position of resonant peaks in certain harmonics, which may result in peaks not being detected using the narrow single frequency window and a fixed frequency resolution. Additionally, certain overtones may have spurious peaks. The intensity of these spurious peaks may be significantly lower than its harmonic peak which may result in spurious peaks not being detected using the single frequency window and a fixed resolution.
Given that certain harmonic peaks and spurious peaks may not be properly detected, a material have a thickness M, may not be properly characterized. For example, different harmonics have different penetration depths in a material such as a film or liquid. By missing one or more harmonics, information regarding its associated visco-elastic properties at this thickness may also be missing. For example, higher crystal harmonics probe closer to the film-crystal-interface, while lower crystal harmonics probe closer to the film-environment interface.
When the acquisition parameters for testing modality may be changed in a characterization system, it generally requires an experienced researcher for selection of reasonable experimental parameters, selection of data processing and analytic tools and approaches, and interpretation of experimental results, which presents a significant limitation, and oftentimes leads to multiple partially finished redundant experiments/tests. For instance, the task of characterizing the changes of a single material functionality (i.e. electrical) in response to environment requires researcher input for instruments and data acquisition (integration time, sampling rates, biases), programming a series of environmental conditions inside the sample chamber (incident light intensity, pressure, temperature, humidity, gas/vapor pressure and composition), and specifying software protocols for data processing and modeling (typically performed after the experiment). However, optimizing the acquisition parameters is difficult, especially for materials which exhibit non-linear, heterogeneous or switching (for instance going from low conductivity state to high conductivity state) responses. Additionally, adjustment of experiment settings on the fly commonly leads to a series of redundant experiments and is prohibitively time-consuming in R&D workflows. Testing is commonly done on different instruments (electro-optical viscoelastic, gravimetric measurements would require 4 different instruments) the data processing routines are difficult/impossible to couple due because of proprietary software used in instruments where the functionality is measured. If the parameter of the modality measurement is selected incorrectly, the entire experiment needs to be redone. If modalities are measured on different instruments, there may be difference between the environments, for instance different size of the environmental compartment which leads to different time for environment stabilization. Such difference will hamper comparison between dynamic response of the materials, and thus loss of important information. Moreover, during measurement of single functionality researcher does cannot access correlation of properties between multimodal response, and thus can not conclude if the test parameters environmental conditions are set correctly. Also, research does not have capability to access if the test is a success. This information becomes available after post-experiment data processing. Thus, the feed back loop part (enabling selection of correct environmental conditions or parameters of the modality test) is delayed by hours or days. Such situation slows the development of novel functionalities and limit testing of novel material compositions.
Moreover, for a known system each modality may require specific format of substrate and film thickness for testing. Optical, electrical, viscoelastic and gravimetric may require up to four separate substrates and four different films.
Accordingly, disclosed is an integrated multifunctional environmental characterization system (IMECS). The IMECS may comprise a memory, one or more interfaces and a processor. The memory may be configured to store one or more machine learned models correlating one or more environmental conditions adjacent a thin film with one or more properties of the thin film. The one or more properties may comprise one or more properties from at least one gravimetric/viscoelastic, electrical or optical properties groups. The processor may be configured to predict an environment condition adjacent to the thin film using the one or more machine learned models from one or more measured properties of the thin film received via the one or more interfaces; and/or predict values for one or more properties of the thin film using the one or more machine learned models from an environmental condition received via one of the one or more interfaces; and display the predicted environment condition and/or the predicted one or more properties.
In an aspect of the disclosure, a sample of the thin film may be deposited on a quartz crystal. Electrodes may be disposed on the quartz crystal to be at least partially covered by the thin film. The electrodes may be for QCM measurements or electrical measurements. The optical measurements may be done in back reflection mode when the probe light is reflected of the surface of QCM electrode (passing through a film twice which leads to improved sensitivity) or in transmittance mode, when the probe light passes through the film of interest and transparent part of the QCM crystal.
Also disclosed is a fluid-flow cell (flow cell). The flow cell may be arranged to encompass the quartz crystal and the thin film. The flow cell may be configured to maintain a controlled environmental condition around the quartz crystal and the thin film and circulate a fluid adjacent to the thin film in conjunction with the one or more environmental control modules. The fluid may comprise gas, vapor, liquid and/or a combination thereof.
In an aspect of the disclosure, the flow cell may enable simultaneous measurement of electrical, optical and gravimetric/viscoelastic properties.
In other aspects, disclosed is an IMECS which may comprise a quartz-crystal microbalance (QCM), electrodes, a user interface, and a processor. The QCM may comprise a quartz crystal having a face onto which a thin film is deposited. The electrodes may be disposed on the face of the quartz crystal to be at least partially covered by the thin film. The user interface may be configured to receive acquisition parameters for two or more characterization modules. The characterization modules may be selected from a group consisting of an optical characterization module, an electrical characterization module and a gravimetric/viscoelastic characterization module. The processor may be configured to adjust the acquisition parameters used to acquire values of one or more properties of the thin film by each respective characterization module from the received acquisition parameters via the user interface based on measured values for the same properties.
The file of this patent contains at least one drawing executed in color. Copies of this patent with color drawing(s) will be provided by the Patent and Trademark Office upon request and payment of the necessary fee.
In accordance with aspects of the disclosure, one or more machine learned models may be deployed to predict one or more properties of a material under certain environmental condition(s) and/or predict an environment condition(s) near the material from a measured or calculated one or more properties of the material. Further in accordance with other aspects of the disclosure, the acquisition parameters for acquire values associated with one or more characteristics (or properties) of the material may be adjusted based on previous measured values associated with the same one or more characteristics (properties)(or calculated).
In an aspect of the disclosure the system may comprise one or more characterization modules 10A-10C (collectively “10”). Only three modules 10 are shown in
In an aspect of the disclosure, the characterization modules 10A-10C may include modules for electrical, optical, gravimetric and viscoelastic characterization.
For example, characterization module 10A may be an electrical characterization module (hereinafter “electrical characterization module 10A”). The electrical characterization module 10A may comprise one or more of a source-meter, a frequency response analyzer (such as an impedance analyzer or function generator), a voltage source, and/or a current source. The electrical characterization module 10A may be used to characterize a DC and/or AC response. In some aspects, the electrical characterization module 10A may measure or calculate impedance spectroscopy, current-voltage (I-V), cyclic-voltammetry (C-V), and transistor measurements, current and voltage (such as pulsed voltammetry).
For example, the source-meter may be a Keithley 240. In an aspect of the disclosure, the impedance analyzer may be a Solarton 1260 available from Ametek® Scientific Instruments.
In an aspect of the disclosure, a maximum sweep voltage may be determined automatically based on measured film current and rate of change of current with respect to applied voltage. A specific capacitance may be calculated using the integral of the C-V curve at each bias sweep rate. For impedance measurements, the frequency range and applied AC amplitude may be determined by measuring sample impedance and signal-to-noise ratio (SNR) in impedance spectra. This measurement may be in a known environment for establishing default values or for calibration. In post-processing routines, electrical impedance data may be fitted to extract distribution of relaxation times (DRT).
The electrical characterization module 10A may connected to a processing system 50 via one or more general purpose interface bus (GPIB) ports (identified in
The optical characterization module 10B may comprise an optical spectrometer and a light source such as but not limited to laser coupled through fiber optic cable. In some aspects, the spectrometer may be obtained from Ocean Optics Inc. as model no. USB 4000 Fiber Optic Spectrometer. The spectrometer may be connected to the processing system 50 via a USB port or a serial port (one of the interfaces 210). The settings used to acquire the optical spectra may comprise, but are not limited to, excitation light intensity, wavelength range (also referred to herein as sampling frequency window or sample frequency window) and signal integration time. In accordance with aspects of the disclosure, the acquisition settings (such as described above) may be controlled autonomously using active feedback from previously measured optical spectra. In one aspect, this control may target a stable optical intensity without saturating the photodetectors in the spectrometer. In an aspect of the disclosure, a target maximum spectral intensity may be between 70 and 80% of photodetector saturation. The adjustment may be iteratively performed to maintain the intensity within the target maximum spectral intensity. For example, when maximum spectral intensity measure is greater than 80% of the photodetector saturation, integration time may be decreased by a preset increment. When maximum spectral intensity is below 70% of photodetector saturation, the integration time may be increased by a preset increment. The preset increment may be 100 μs. However, other increments may be used. If the maximum spectral intensity is still outside the target maximum spectral intensity, the integration time may be adjusted again.
The gravimetric and viscoelastic characterization module 10C may comprise a portable antenna analyzer coupled to electrodes. For example, the portable antenna analyzer/network analyzer may be a SARK-110. The portable antenna analyzer may be used to measure conductance and/or impedance. The portable antenna analyzer may be connected to the processing system 50 via a USB port (one of the interfaces 210) or wirelessly WIFI/BLUETOOTH (registered Trademark). The portable antenna analyzer may also be electrically connected with electrodes.
Each of the respective devices in the characterization modules 10A-10C may be controlled by the processing system 50. A block diagram of the processing system 50 is shown in
The interfaces 210 may be USB interfaces, GPIB interfaces, serial interfaces, . . . etc. In some aspects, the devices in the characterization modules 10 may communicate with the processing system 50 wirelessly.
The combination of the multiple characterization modules 10 enables multi-functional characterization of a material (such as a film) across an ultra-broadband frequency range as shown in
The film 410 may cover the central electrode 405 (used for QCM) and spans the quartz crystal 425 to contact the electrodes 400 (allowing in-plane electrical measurements) as well as the QCM measurements. The electrical characterization module 10A may be connected to the electrodes 400 using electrical leads and paste. As shown, a flow cell cover 420 may have an opening 415 to able light to pass or a fiber optic cable to pass to enabled optical measurement via the optical characterization module 10B. In some aspects, the light may be a halogen lamp. In other aspects, the fiber optic cable may run into the flow cell or be in contact with glass quartz (optical window) as described herein. Optical window may also allow for sample heating and surface activation by infrared (IR) and ultraviolet (UV) light sources mounted outside the chamber
In some aspects of the disclosure, additional characterization modules 10N may be used such as photoluminescence (PL 432), Ramon spectroscopy 434, scanning probe microscopy 436, 438 such as atomic force microscope (AFM). The cantilever 430 for the AFM is shown in
In accordance with aspects of the disclosure, the processing system 50 may control the environment around the sample (sample environment 450) via one or more environmental control modules 20A-20C. In aspect of the disclosure, one of the environmental control modules (e.g., 20A) may be a humidity generator to control the relative humidity. In an aspect of the disclosure, RH-200 humidity generator (L&C Science) may be used. The humidity generator may be connected to the processing system 50 via a USB (one of the interfaces 210). In other aspects, one of the environmental control modules (e.g., 20B) may control gas/vapor flow for different types of gas. In this aspect, the environmental control module 20B may comprise a mass flow controller such as from Alicat Scientific. The mass flow controller may be connected to the processing system 50 via an RS-232 port (one of the interfaces 210). In another aspects, one of the environmental control modules (e.g., 20C) may control the pressure. In this aspect of the disclosure, the environmental control module 20C may comprise a pressure transducer and a flow valve. For example, the pressure may be controlled using a 651C Digital-Analog Pressure Controller with 972 DualMag pressure transducer and butterfly valve (MKS Instruments).
In other aspects, other environmental control modules may be used such as temperature and light. For example, the temperature may be controlled a precision temperature controller such as Model 250 precision temperature controller (J-KEM Scientific).
In accordance with aspects of the disclosure, the processing system 50 may dynamically control the sample environment 450 using a user-defined sequence or automatically. The user-defined sequence may control temperature, pressure, humidity, gas/vapor concentration, and light intensity for a user-defined period of time, ranging from seconds to days, depending on the kinetics of material response to dynamic environmental changes. The user-defined sequence may be input using a user interface 55. An example of a user interface 55 is shown in
The user may also use the user interface 55 to input one or more acquisition parameters for the characterization modules 10. In an aspect of the disclosure, the user interface 55 may comprises a plurality of separate portions for the inputs for electrical acquisition parameters (portions 525), the inputs for optical acquisition parameters (portion 520) and inputs for QCM acquisition parameters (portion 515).
For example, in an aspect of the disclosure, the system may acquire a QCM conductance/impedance at a fundamental resonance and a plurality of harmonics. The number of harmonics may be selected based on the application and film 410. For example, the system may acquire the conductance/impedance at odd harmonics. The odd harmonics may be from the 3rd to 17th harmonic (where n represents the harmonic number). In an aspect of the disclosure, the user may define a sample frequency window for the fundamental resonance and each harmonic. For example, the user may input a center frequency for the sample frequency window and a frequency width (e.g., +/a frequency). In an aspect of the disclosure, the width increases for larger harmonics. For example, the width is 8 KHz for n=7 and 18 Khz for n=17. The user may also input the resolution. The user may also input other acquisition parameters for QCM-D such as a number of points to average at each frequency. As the number of points increase, the SNR ratio is increased. In some aspects, certain harmonics (higher harmonics) may be ignored when the SNR is below a threshold.
In an aspect of the disclosure, prior to depositing a material on the quartz crystal 425, the conductance spectra of the quartz crystal may be obtained to confirm the viability of the crystal for characterizing a film 410. The fundamental resonance of the quartz crystal is identified by a manufacturer of the quartz crystal. For example, the fundamental resonance may be 5 MHz. Harmonics of the quartz crystal may also be obtained. The same harmonics may be obtained as will be used in the characterization session such as an experiment. The measurements may be done in air. The acquisition parameters used for this initial characterization may also be used as the initial acquisition parameters once the material such as a film 410 is deposited on the quartz crystal 425.
The user may use the portions 525 to input the electrical acquisition parameters for both DC and AC measurements. For example, for the DC measurements, the user may input the bias, the maximum sweep bias, voltage steps and C-V sweep rate among other parameters. For AC measurements such as impedance and phase, the user may input the start and end frequency. Additional AC acquisition parameters may be the AC amplitude, the DC offset and number of signals to average.
The user may use the portions 520 to input the optical acquisition parameters such as the integration time. Although not specifically shown in
The user interface 55 may be co-located with the processing system 50. For example, the processing system 50 optionally may include a display. The user interface 55 may be displayed on the display. In other aspects, the user interface 55 may be installed in another device such as a personal computer, mobile terminal such as a mobile telephone, smartphone or laptop, etc.
In accordance with aspects of the disclosure, the acquisition parameters may be adjusted in real-time based on measured/calculated characteristics of a material such as a film.
The housing 610 may also have slots 620 configured to receive connectors 600, 605. One connector is for QCM-D measurement 600 and another connector is for electrical measurements 605. In an aspect of the disclosure, the connectors 600, 605 may be BNC connectors. The BNC connectors are inserted into the slots 620. The slots 620 may be dimensioned be maintain a snug fit with the connectors 600, 605 such that the environment within the flow cell 600 is unaffected by the external environmental conditions.
The housing 610 and cover 420 form a compartment for the sample 30. In an aspect of the disclosure, the sample 30 may comprises the quartz crystal 425 and film 410 (under measurement) and the electrodes 400, 405. The sample 30 may be held and suspended in air by a sample holder 625. The holder 625 prevents the sample 30 from contacting the housing 610. The holder 625 is connected to one of the connectors 600, 605. As shown in
The flow cell cover 420 may comprise an opening. The opening 415 may be filled with a glass quartz. The opening 415 (with glass quartz) enables optical measurements 310 and other spectroscopy measurements.
As shown in
In a case where the optical measurements 310 use the transmittance 642 (where a film is transparent to the emitted wavelengths), the housing 610 may have another opening (not shown in
As described above, the electrical 300, optical 310 and QCM-D 305 measurements may be made at the same time. In this aspect, the electrical measurements may be made using the electrode 400A/400B at the edges of the film 410/quartz crystal, the QCM-D measurements 305 may be made using the electrodes 405 at the middle of the film 410/quartz crystal and the optical measurements 310 may be made at a location between the electrodes 405, 400A/400B. Since the measurements are made at different positions, the measurements of one type of property may not impact the measurements of another type of property. For example, since the electrical measurements 300 are made at the edges and the edges do not vibrate, the electrical measurements do not impact the QCM-D measurements 305. The flow cell 600 may be used in a gas or vapor controlled environment, however, when the environment includes liquid, the wires, connectors and holder may be covered with material to provide resistance to damage by the liquids.
In other aspects of the disclosure, a different flow cell 600A may be used when the environment may include liquid flow such as shown in
The opening 415 may be filled with glass quartz. The opening 415 (with the glass quartz) may be used for optical measurements 310. In an aspect of the disclosure, a fiber optic cable may be positioned in contact with the glass quartz for emitting and detecting wavelengths to/from the sample 30.
In other aspects of the disclosure, a flow cell 600B may be used for characterization of a sample 30 with respect to a Reference 30A. As noted above, the sample 30 may include the film 410 and quartz crystal 425. However, the reference may include the quartz crystal 425 without the film 410. The electrodes 400 may be deposited on the edges of the quartz crystal as shown in
In an aspect of the disclosure, the housing 610A may having an opening 415 on the top. This opening 415 may be filled with glass quartz. The opening 415 (with the glass quartz) may be used for optical measurements 310. In an aspect of the disclosure, a fiber optic cable 2400 may be positioned in contact with the glass quartz for emitting and detecting wavelengths to/from the sample 30. In an aspect of the disclosure, the reference 30A may be positioned below the sample 30 and aligned. Both may be parallel to the opening 415. In other aspects, the glass quartz may be omitted and the fiber optic cable 2400 may be inserted into the opening 415 and positioned adjacent the sample 30. As shown in
A reference 30A may be used when the sample 30 is composed of a composite material such as having a nanomaterial. The reference 30 may be a polymer material (which is part of the composite material). In this case, a QCM-D response of the reference 30 may be subtracted from the QCM-D response of the composite material.
In other aspects of the disclosure, the same flow cell 600C may be used to characterize the properties of multiple films 410 at the same time.
In some aspects, multiple characterization modules may be controlled to operate at the same time to characterize different properties of the material simultaneously. For example, in some aspects, the electrical characterization module 10A and the gravimetric/viscoeleastic characterization module 10C may acquire property values at the same time. In other aspects, electrical characterization module 10A, the optical characterization module 10B and the gravimetric/viscoeleastic characterization module 10C may acquire property values at the same time. In other aspects, the optical characterization module 10B and the gravimetric/viscoeleastic characterization module 10C may acquire property values at the same time. In other aspects, each characterization module 10 may acquire property values at different times. The acquisition of the values may be in one or more different environments. Similar to the characterization modules, the environmental control modules 20 may be independently operated only or in any combination.
The design characterization process (e.g., design of the experiment) may comprise starting one or more programs to execute the functionality described herein and choosing one or more storage locations for the results at S700-1. In an aspect of the disclosure, the acquired/calculated properties may be locally stored. In other aspects, the acquired/calculated properties may be transmitted to a server for storage, and in some aspects, for further processing.
At S700-2, the user may use the user interface 55 to select the relevant modules (e.g., characterization modules 10 and/or environmental control modules 20). For example, the system control portion 510 may have a check box or drop down menu for the selection of the different modules. Different combinations of characterization modules 10 and/or environmental control modules 20) may be selected. For example, the user may select the relative humidity control module (which may include RH-200) in combination with the gravimetric and viscoelastic characterization module 10C (which may include the SARK-110) and electrical characterization module 10A (which may include the Solartron 1260). In the example, in
At S700-3 and S700-5, the user may use the user interface 55 to select the relevant material properties and input certain acquisition parameters. In an aspect of the disclosure, the acquisition parameters may be mode specific. There may be two modes of acquisition, manual and automatic. In manual mode, the user may set all for the acquisition parameters in advance for each iteration of acquisition. In automatic mode, the user may set the initial acquisition parameters and a sequence of acquisition environments and the processor 200 may calculate successive acquisition parameters for each iteration using the measured values of the material properties (calculated values). The user may set the mode in S700-4 using the user interface 55.
The following description will be made with respect to a selection of the relative humidity environmental control module (including the RH-200). However, this description equally applies to the other environment control modules.
At S700-5, the user may set the sequence of relative humidity used to acquire the material properties. The sequence may include the specific humidity level and the time spent at each humidity level. For example, the setting may include humidity levels every 20% between 20%-80% and a low humidity such as 2% and a high humidity of 95% as shown in
At S700-5, the user may set the initial acquisition parameters for the gravimetric and viscoelastic characterization module 10C. In aspect of the disclosure, the acquisition parameters may include the number of harmonics, the sampling frequency windows (center frequency and width) and resolution and multiple harmonics for spatial information. For example, highest crystal harmonics correspond to regions closer to the film-crystal interface and lower harmonics correspond to regions closer to the film-environment interface as shown in
Similarly, at S700-5, the user may set the initial acquisition parameters for the other characterization modules 10A/10B.
Once all of the initial acquisition parameters are set for each characterization module 10 and the environmental control modules 20, the processor 200 may transmit the settings to the respective equipment at S705-1-S705-4. For example, the processor 200 may transmit the set relative humidity to the RH-200, the set sample frequency windows (set of) to the SARK-110, the set frequency ranges and/or bias, step size to the Solartron 1260 and/or Keithley (electrical characterization module 10A) and the optical frequency range, resolution and integrating time to the Ocean Optics (optical characterization module 10B).
At S710-1-S710-3, the respective devices obtain the values for the respective film properties under the set humidity level (e.g., first humidity level). The RH-200 maintains the humidity level at the set point at S710-4. Specifically, the RH-200 causes the set humidity to flow into the flow cell (e.g., 600). At S720-4, the processor 200 determines whether the material response has stabilized, e.g., are the properties changing. When the properties have stabilized, the processor 200 obtains the next environmental condition based on the sequence at S715-4 and controls the RH-200 to produce the next relative humidity at S705-4.
For the gravimetric and viscoelastic characterization module 10C (which may include the SARK-110), the SARK-110 measures the QCM conductance spectra using the sweeping frequencies within the sample frequency window set (plurality of sample frequency windows) received from the processor 200 at a received resolution. In S720-3, the SARK-110 may deconvolute and determine the peaks (e.g., fit peaks). Each measured QCM conductance spectrum may be automatically fitted to a Butterworth-Van Dyke (BVD) lumped-element equivalent circuit model. Information extracted from the BvD model may include frequency shift (change in the resonant peak position Δf) and dissipation change (change in the value of full width at half max of the resonance peak ΔD). Parameters of the fit may be used as initial conditions for the following fit to ensure that initial fitting parameters are close to the convergent parameters. Sometimes the fitting procedure may fail to find a fit because an initial guess for the fitting parameters is too distance from the correct parameters. To alleviate this, best guess for the initial fitting parameter values for the first set of spectra is used, but for subsequent cycles (values), the values for the previous cycle are used as the initial guess. Further, to avoid fitting of distorted resonant peaks, undesirable spurious overtone peaks may be fitted and removed from the resonance peak by subtraction.
Δf and ΔD may be used to estimate changes in elastic modulus and viscosity using one or more models (such as the continuum mechanics viscoelastic model). The calculated properties and fitted peaks may be transmitted to the processing system 50. The processor 200 may store the peaks and calculated properties in the memory 205 in association with the set humidity level. The processor 200 may also display the calculated properties and the recorded conductance spectra. In some aspects, the output may be displayed on the user interface 55 (in the output portion 530).
In other aspects of the disclosure, instead of the gravimetric and viscoelastic characterization module 10C (which may include the SARK-110) performing S720-3, the module 10C may transmit the raw data to the processing system 50 for processing.
At S715-3, the processor 200 may adjust one or more of the acquisition parameters for the QCM-D measurements. In accordance with aspects of the disclosure, dynamic sampling frequency windows 900 are used in each iteration. The dynamic sampling frequency windows 900 track the shifts in the QCM resonance peak(s). The tracking causes the windows 900 to shift to automatically center the respective sampling frequency window(s) 900 on the resonance peak position(s).
The width (e.g., bandwidth) of each sampling frequency window 900 in the window set are iteratively adjusted to accommodate changes in the peak width, ensuring that peak shoulders do not stretch outside the measured frequency range. In QCM spectra which contain spurious overtone peaks adjacent to the main resonant peak, the automated detection of peaks, tracking of peak positions, and prediction of future peak positions may be performed.
In an aspect of the disclosure, the tracking and prediction of future peak positions may be performed using machine learning such as a support vector regression. In some aspects, the SVR may be implemented with Scikit-library in Python. As the number if iterations increase, the model becomes more accurate.
In an aspect of the disclosure, the predicted peak positions may be used as a reference for optimizing an initial guess for values of peak parameters during peak fitting and spectral deconvolution. The initial peak fit parameters may be adjusted iteratively as new spectra are acquired and tracked.
In an aspect of the disclosure, a single peak may be separately fitted to avoid interfering from the adjacent spurious peaks (after deconvolution of peaks of interest).
In accordance with aspects of the disclosure, the same model that is used to predict the peaks, may be used to predict a frequency shift under a different environmental condition (e.g., different humidity level). In other aspect, the data (peak fit parameters as a function of environmental condition) may be used for the calculation of a different regression model. Integration of regression fitting during the active measurements enabled prediction of material response during the next programmed environmental conditions by extrapolation of regression beyond already measured conditions. The regression model may also allow further refinement of the measurement settings, fitting parameters, and sampling window.
Before transitioning to the next measurement condition (e.g., humidity level), the processing system 50 may assess material response stability by performing event detection (ED) analysis. The ED analysis is based on monitoring changes in moving averages and variance of the Δf response.
Once the acquisition parameters are adjusted by the processor 200, the processor transmits the new parameter to the gravimetric and viscoelastic characterization module 10C (which may include the SARK-110) to use for the next measurements. S705-3, S710-3, S720-3 and S715-3 may be iteratively repeated for each environmental setting.
At S725, the processor 200 may generate and display one or more reports for the QCM-D measurements.
The processor 200 may control the optical characterization module 10B similar to the gravimetric and viscoelastic characterization module 10C. For example, at S705-2, the processor 200 may transmit to the optical characterization module 10B (which may include the Fiber Optic Spectrometer) the optical sample frequency window, the resolution and the integration time. At S710-2, the Fiber Optic Spectrometer may transmit light within the optical sample frequency window and measure the films 410 transmittance or reflectance spectra. At S720-2, the Fiber Optic Spectrometer deconvolute the measured data and determine the peaks. The spectrometer may also calculate the SNR. The deconvolution and peak fitting may be similar to described above, where each peak may be fit to a Gaussian distribution function and the peak position, max height, and width (FWHM) may be extracted. Default values for the initial fitting parameters for the first fit may be used, and at subsequent cycles of measurements of the optical spectra, calculated fit parameters of the previous spectrum may be used as an initial guess for the fit parameters of the current spectrum.
The calculated properties (such as SNR) and fitted peaks may be transmitted to the processing system 50. The processor 200 may store the peaks and calculated properties in the memory 205 in association with the set humidity level. The processor 200 may also display the calculated properties. In some aspects, the output may be displayed on the user interface 55 (in the output portion 530).
In other aspects of the disclosure, instead of the optical characterization module 10B (which may include the Fiber Optic Spectrometer) performing S720-2, the module 10B may transmit the raw data to the processing system 50 for processing.
At S715-2, the processor 200 may adjust one or more of the acquisition parameters for the optical spectra acquisition. In accordance with aspects of the disclosure, dynamic sample frequency windows are used in each iteration. The dynamic sampling frequency windows track the shifts in the optical spectra peaks. The tracking causes the windows to shift to automatically center the respective sampling frequency window(s) on the peak position(s).
In some aspects, the width (e.g., bandwidth) of the optical frequency window may be iteratively adjusted to accommodate changes in the peak width. In an aspect of the disclosure, the peaks may be detected, tracked and used to predict future peak positions. In an aspect of the disclosure, the tracking and prediction of future peak positions may be performed using machine learning such as a support vector regression. In some aspects, the SVR may be implemented with Scikit-library in Python. In an aspect of the disclosure, the predicted peak positions may be used as a reference for optimizing an initial guess for values of peak parameters during peak fitting and spectral deconvolution. The initial peak fit parameters may be adjusted iteratively as new spectra are acquired and tracked.
Also, in accordance with aspects of the disclosure, the same model that is used to predict the peaks, may be used to predict a reflection/transmittance under a different environmental condition (e.g., different humidity level). In other aspect, the data (peak fit parameters as a function of environmental condition) may be used for the calculation of a different regression model. Integration of regression fitting during the active measurements enabled prediction of material response during the next programmed environmental conditions by extrapolation of regression beyond already measured conditions. The regression model may also allow further refinement of the measurement settings, fitting parameters, and optical frequency sampling window.
Further, as described above, the integration time may be adjusted iteratively to maintain the intensity with the target maximum spectral intensity. This adjustment may be made by the processor 200 based on the calculated SNR. A low SNR may increase the integration time whereas a high SNR may reduce the integration time.
Once the acquisition parameters are adjusted by the processor 200 (e.g., optical frequency sample window and integration time), the processor 200 may transmit the new acquisition parameters to the optical characterization module 10B (which may include the Fiber Optic Spectrometer) to use for the next measurements. S705-2, S710-2, S720-2 and S715-2 may be iteratively repeated for each environmental setting.
At S725, the processor 200 may generate and display one or more reports for the optical measurements.
The processor 200 may control the electrical characterization module 10A similar to the other modules. For example, when the measured property is an impedance, the processor 200 may transmit the initial acquisition parameters to an impedance analyzer such as the Solartron 1260 at S705-1. The initial acquisition parameters may include the frequency range (start and end) and the number of point (step size). The initial acquisition parameters may also include an AC amplitude and offset (DC). At S710-1, the impedance analyzer may use the initial acquisition parameters to measure the impedance of the film 410 under the first environmental condition, e.g., 2% relative humidity. The impedance analyzer may transmit the measured impedance over the frequency range to the processing system 50. At S715-1, the processor 200 may determine whether to adjust the acquisition parameters based on the measured impedance. In an aspect of the disclosure, the tracking of the impedance and prediction of future impedance may be performed using machine learning such as a support vector regression. For example, the machine learning model may be trained using the samples responses measured at an initial set of environmental conditions, for example, the impedance measured at 2%, 4% and 6% relative humidity. The trained model may then be used to predict the expected impedance of the sample, at e.g., any relative humidity value inside the range from 0-100% relative humidity.
Of note, the low frequency impedance provides good characterization information of the film 410. Therefore, in an aspect of the disclosure, the start frequency of the frequency sweep for the impedance and step size may be adjusted based on a prior measurement (previous iterations).
Once the acquisition parameters are adjusted by the processor 200 (e.g., start frequency and step size) the processor 200 may transmit the new acquisition parameters to the impedance analyzer to use for the next measurements. S705-1, S710-1, and S715-1 may be iteratively repeated for each environmental setting.
At the same time of the impedance measurement, other electrical properties of a film 410 may also be acquired. In other aspects, the other electrical properties of the film 410 may be separately acquired. For example, cyclic-voltammetry (C-V) and current-voltage (I-V) responses may be acquired. The response may be measured by a source-meter. As S705-1, the processor 200 may transmit the initial acquisition parameters to the source-meter. The initial acquisition parameters may include the sweep rate. In an aspect of the disclosure, multiple sweep rates may be used. For example, three different sweep rates may be used. The number of different sweep rates is not limited to three and other different sweep rates may be used. The sweep rates may be 25, 50 and 200 mV/s. However, the sweep rates are not limited to the listed rates such as shown in
At S710-1, the source-meter may measure the current through the film at the first environmental condition (e.g., 2% relative humidity) for the three different sweep rates and different bias within the maximum bias. The current may be measured via the electrodes 400.
At S720-1, the source-meter may transmit the measured current along with information of the acquisition parameters to the processing system 50 for processing. At S720-1, the processor 200 may calculate the capacitance using the measured current and bias voltage.
At S715-1, the processor 200 may adjust the acquisition parameters for the next iteration based on the measured current or the calculated capacitance. For example, the processor 200 may change the maximum bias based on the measured current or the calculated capacitance. In an aspect of the disclosure, the tracking of the current, capacitance and prediction of future current and capacitance may be performed using machine learning such as a support vector regression. For example, the machine learning model may be trained using the sample responses measured at an initial set of environmental conditions, for example, the capacitance or current measured at 2%, 4% and 6% relative humidity. The trained model may then be used to predict the expected capacitance of the sample or current flowing in the sample at e.g., any relative humidity value inside the range from 0-100% relative humidity. In an aspect of the disclosure, the max bias may be modified in order to increase/decrease the resultant current. For example, when a measured current is high, the max bias may be reduced and vice versa.
Measured current may increase as relative humidity increases from 2% to 95%, which is consistent with results from impedance spectroscopy.
At S725, the processor 200 may display results of the electric characterization on a display such as described above, e.g.,
Other electrical properties of the film 410 may include a DC electrical response (e.g., current). The acquisition parameter for the DC electrical response may be a DC bias (voltage). The processor 200 at S705-1 may transmit to the source-meter the set DC bias. The set DC bias may be defined in S700-5 by the user. For example, the DC bias may be 10 mV. However, other bias may be used. At S710-1, the current is detected by the source-meter (as a function of the relative humidity). The same relative humidity sequence may be used as described above. The current may be measured after the environment stabilizes, e.g., reaches the target humidity and remains for a set time.
At S720-1, the processor 200 may calculate an exponential decay model to examine kinetics of the change in current at each relative humidity step. In an aspect of the disclosure, the current response may be automatically fitted to an exponential decay model. This is shown the center of
In an aspect of the disclosure, the automated analysis of the DC electrical current response to changing relative humidity enables rapid extraction of electrical hysteresis information (upper left inset), kinetics of the response (upper right inset), and dependence of electrical conductivity behavior on ambient environment. In an aspect of the disclosure, the red fit lines shown in the center panel may be used to determine the rate(s) of change of the measured current. For example, when the current is changing, the environment, e.g., relative humidity, may be held constant so that the sample response may stabilize. Once the current reaches a stable value and does not change over time, a signal may be sent from the processor to the environmental control module to change the environment, e.g., increase or decrease the relative humidity.
In the above description examples, the film 410 contained PEDOT:PSS.
As described above, one or more machine learned models may be deployed to predict one or more properties of a material under certain environmental condition(s) and/or predict an environment condition near the material from a measured or sensed one or more properties of the material.
Of note, one or more measured or calculated film properties may be correlated with one or more other measured or calculated film properties in different environmental conditions. These correlations enable the properties to be predicted using one or more machine learned models.
In an aspect of the disclosure, the correlations may be quantified using a parameter such as a bivariate (Pearson) correlation coefficient. The coefficient quantifies the relationships between gravimetric/viscoelastic, electrical, and optical humidity properties of a film in response to the different environment conditions (such as relative humidity).
The machine learned model may be a least absolute shrinkage and selection operator (LASSO) linear regression. LASSO regression may be used as a predictive model because it eliminates insignificant regression coefficients to improve interpretability of the regression, enabling simple comparison between the importance of each feature in the model.
In all cases, supplementing training data with data from other measured modalities may significantly improve accuracy of the model prediction. This example sheds light on the benefits of supplementing measured data with other auxiliary features for improving predictive modeling.
The accuracy of any machine learned model (using LASSO) described above, depends strongly on the proportion of measured data which is used for training compared to that used for testing.
When 2% of the data may be used for training, the training set may consist of only 4 sets of data (also referred to herein as samples): features measured at e.g., 2%, 32%, 62%, and 92% relative humidity. However, when 50% of the sets of data may be used for training, the training set may consist of one training set for each integer relative humidity value and one testing set for each half-integer relative humidity value. As the percentage of sets of data used for training increases from 2% to 50%, error of the predictive model decreases as expected in the example depicted in
Once again, in all cases in the example, errors are lower when a single functional modality is supplemented with additional functional modalities. For example, when all modalities (electrical, optical, gravimetric/viscoelastic) are used to predict an optical modality (“All to O” in plot legend), error is significantly lower than when only optical modalities are used to predict optical modalities (“O to O” in legend). As shown the blue curve with the open circles is the error in prediction with only optical modalities (error >4%) whereas the blue curve with blue solid circles is the error in prediction with all modalities for the optical modalities (error <2%).
The enhancement in predictive accuracy is shown in the bottom of
In an aspect of the disclosure, a plurality of models may be trained with 4% of the data (96% may be used for testing). The difference between the environmental conditions (such as relative humidity between the data points used in training may be the same, e.g., every 10% or every 20%. In other aspects, the difference may be non-linear. The specific percentages used in training may be user defined or based on apriori knowledge of the equipment.
Model testing may be accomplished using 5-fold cross-validation, where 4% of the data set is randomly selected to be used for training, and the remaining 96% of the dataset is used for testing. This process is performed for each model type and each combination of hyperparameters, and repeated 5 times so that a different training dataset is selected each time.
The regression models may include multiple different ML techniques such as support vector regressions, neural networks, ensemble methods, linear methods and tree-based methods. A non-exhaustive list of model algorithms includes automatic relevance determination regression (ardregression), degree-1 polynomial fit (poly1), degree-2 polynomial fit (po1y2), degree-3 polynomial fit (poly3), adaboost decision tree regressor (adaboostregressor), bagging decision tree regressor (baggingregressor), bayesian ridge regression (bayesianridge), elastic net regressor (elasticnet), Huber regressor (huberregrressor), least-angle regression (lars), cross-validated least-angle regression (larscv), lasso regression (lasso), cross-validated lasso regression (lassocv), lasso model fit with least-angle regression (lassolars), cross-validated lasso model fir with least-angle regression (lassolarscv), lasso model fit with least-angle regression using information criterion (lassolarsic), linear regression (linearregression), orthogonal matching pursuit model (orthogonalmatchingpursuit), cross-validated orthogonal matching pursuit model (orthogonalmatchingpursuitcv), passive aggressive regressor (passiveagressiveregressor), ridge regression (ridge), cross-validated ridge regression (ridgecv), stochastic gradient descent regressor (sgdregressor), Thiel-Sen regression (thielsenregressor), decision tree (decisiontreeregressor), random forest (randomforestregressor), extra tree regressor (extratreeregressor), support vector regression with linear kernel (linearsvr), nu support vector regression (nusvr), and support vector regression with radial basis kernel (svr).
The number of different models trained may be set by a user. For example, more than 10 models may be trained, more than 20 models may be trained, more than 30 models may be trained. Each model may be trained using different hyperparameters.
For purposes of the description only, material properties measured at e.g., 2, 10, 20, 50, 80, 85, 90, and 95% relative humidity may be initially used for training a plurality of machine learning regression models for fitting nonlinear response to humidity at S1810.
At S1815, the trained models with a given hyperparameter sets are tested to predict a material response at different relative humidity. In an aspect of the disclosure, the different relative humidity may be from 2% on the low end to 96% at the upper end. The interval between predictions may be 0.5% relative humidity. However, in other aspects of the disclosure, different end points and intervals may be used. The processor 200 may compare the predicted values for the property based on the model(s) with the actual measured value/calculated value for the same property to obtain an error percentage.
At S1820, certain models may be selected for different properties based on their performance, e.g., error. For example, the model type with a given hyperparameter configuration which performs the best at predicting the test dataset on average over all the cross-validation splits may be selected to be deployed.
The red in
At S1825, each selected model may be is stored in memory 205 with its associated property(ies). In an aspect of the disclosure, when more than one model has a similar prediction accuracy for a particular property, the model that has the highest prediction accuracy for more properties may be selected and stored. In accordance with aspects of the disclosure, deploying models as described herein may achieve a prediction error of under 7% for all material properties and a mean error of ˜3% across all properties. This is because a broad range of models and hyperparameters may be screened as described above.
As described above, one or more machine learned models may be deployed to predict one or more properties of a material under certain environmental condition(s) and/or predict an environment condition near the material from a measured (calculated) one or more properties of the material.
At S2005, the processor 200 receives target or desired properties for prediction. In an aspect of the disclosure, the user may input one or more properties of the material (such as a film 410) via the user interface 55. The properties may be optical, electrical and/or gravimetric/viscoelastic properties.
At S2010, the processor 200 retrieves one or more stored machine learned models based on the input one or more properties. As noted above, the processor 200 may select the best model for specific properties and associate the model with the properties. Thus, at S2010, the processor 200 may use the inputted one or more properties as the key for selection.
At S2015, the processor 200 uses the retrieved one or more machine learned models to predict the inputted one or more properties using the received environmental condition as the input to the models.
At S2020, the processor 200 outputs the predicted values for the one or more properties. In some aspects, the processor 200 may cause the predicted values for the one or more properties to be displayed on the user interface 55. In other aspects, the processor may transmit the predicted values for the one or more properties to a mobile device.
In other aspects of the disclosure, in addition to outputting the predicted values for the properties, the predicted values may be used to determine the acquisition parameters for the same properties.
At S2100, the processor 200 may determine the acquisition parameters for the one or more properties using the predicted values for the one or more properties. For example, in a case where the properties are a QCM spectra and specifically the fundamental resonance (e.g., Δf1) and certain harmonics with its overtones (e.g., Δf3, Δf5, Δf7, Δf9, Δf11, Δf13, Δf15 Δf17), the processor 200 may determine the sample frequency windows such that each are centered at the predicted frequency, respectively (the fundamental resonance and each respective harmonics). The processor 200 may determine the width of the sample frequency windows such that it includes the predicted frequencies of the overtones.
The processor 200 may transmit the determined acquisition parameters to the respective characterization modules 10. Additionally, the processor 200 may transmit the received environmental condition in S2000 to the environmental control modules (associated with the condition) to maintain the environment at the specific environmental condition. For example, the processor 200 may transmit a target relative humidity to the RH-200.
The characterization modules 10 may obtain the measurements in a similar manner as described above using the received acquisition parameters and may calculate other properties as needed. The characterization modules 10 may transmit the actual measured properties (values) and calculated properties (values) to the processing system 50. At S2105, the processor 200 receives the actual measured properties (values) and calculated properties (values) and may compare the received values with the predicted values for the same properties (which is stored in the memory 205) to determine a percent error for each property (value). The received values are also stored in the memory 205.
Once the actual percent error is determined for each property, the processor 200 may retrieve the expected percent error (predicted) associated with a respective model used for each property. For example, if the Huberregression model was used to predict the fundamental resonance (frequency shift) (Δf1), the processor may retrieve the expected percent error for the same (an example of which is shown in
Similarly, when the Ardreggression model is used, for example, to predict Δμ11 (11th harmonic), the processor 200 may retrieve the expected percent error of (e.g., about 1%) and compare with the actual percent error determined from the calculated value (from the actual measurements).
In other aspects of the disclosure, the user may set an error tolerance. For example, the user may set a 1% tolerance such that if the difference is less than 1%, the alert is not sent, but if the difference is greater than or equal to the tolerance %, the alert is sent.
If at S2110, the processor 200 determines that the actual percent error is less than or equal to the expected percent error, the processor 20 may cause the measured/calculated properties (values) to be displayed at S2115. The values may be displayed on the user interface 55.
The processor 200 may then calculate new acquisition parameter based on the actual measured/calculated properties as described above (and transmit the same to the characterization modules 10).
In an aspect of the disclosure, when the actual percent error is greater than the expected percent error (and the alert is sent), the processor 200 may stop the acquisition process. This may allow the user to check on the film 410/quartz crystal 425 (and other equipment).
At S2010, the processor 200 may retrieve from the memory 205 one or more models associated with both the inputted target environmental condition(s) and the received measured/calculated properties. As noted above, the processor 200 may select the best model (lowest expected percent error) for the received one or more properties (values). The processor 200 may use the received one or more properties as the key for selection (and environmental condition(s)).
At S2205, the processor 200 may use the retrieved one or more machine learned models to predict the inputted one or more environmental conditions using the received one or more properties (values) as the input to the models. For example, the processor 200 may predict the relative humidity around the film 410/quartz crystal 425.
At S2210, the processor 200 outputs the predicted values for the one or more environmental conditions, respectively. In some aspects, the processor 200 may cause the predicted values for the one or more environmental conditions to be displayed on the user interface 55. In other aspects, the processor 200 may transmit the predicted values for the one or more environmental conditions to a mobile device.
As used herein terms such as “a”, “an” and “the” are not intended to refer to only a singular entity, but include the general class of which a specific example may be used for illustration.
As used herein, terms defined in the singular are intended to include those terms defined in the plural and vice versa.
References in the specification to “one aspect”, “certain aspects”, “some aspects” or “an aspect”, indicate that the aspect(s) described may include a particular feature or characteristic, but every aspect may not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same aspect.
Various aspects of the present disclosure may be embodied as a program, software, or computer instructions embodied or stored in a computer or machine usable or readable medium, or a group of media which causes the computer or machine to perform the steps of the method when executed on the computer, processor, and/or machine. A program storage device readable by a machine, e.g., a computer readable medium, tangibly embodying a program of instructions executable by the machine to perform various functionalities and methods described in the present disclosure is also provided, e.g., a computer program product.
The computer readable medium could be a computer readable storage device or a computer readable signal medium. A computer readable storage device may be, for example, a magnetic, optical, electronic, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing; however, the computer readable storage device is not limited to these examples except a computer readable storage device excludes computer readable signal medium. Additional examples of the computer readable storage device can include: a portable computer diskette, a hard disk, a magnetic storage device, a portable compact disc read-only memory (CD-ROM), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical storage device, or any appropriate combination of the foregoing; however, the computer readable storage device is also not limited to these examples. Any tangible medium that can contain, or store, a program for use by or in connection with an instruction execution system, apparatus, or device could be a computer readable storage device.
A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, such as, but not limited to, in baseband or as part of a carrier wave. A propagated signal may take any of a plurality of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may be any computer readable medium (exclusive of computer readable storage device) that can communicate, propagate, or transport a program for use by or in connection with a system, apparatus, or device. Program code embodied on a computer readable signal medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
The terminology used herein is for the purpose of describing particular aspects only and is not intended to be limiting the scope of the disclosure and is not intended to be exhaustive. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the disclosure.
This application claims the benefit of and priority to U.S. Provisional Application Ser. No. 63/081,959 filed on Sep. 23, 2020 and U.S. Provisional Application Ser. No. 63/081,962 filed on Sep. 23, 2020, the entirety of which are incorporated by reference.
The United States Government has rights in this invention pursuant to contract no. DE-AC05-00OR22725 between the United States Department of Energy and UT-Battelle, LLC.
Number | Date | Country | |
---|---|---|---|
63081959 | Sep 2020 | US | |
63081962 | Sep 2020 | US |