Analog Input Research Articles

Design and evaluation of a gas, temperature profiling and data acquisition system to monitor the performance of a batch fixed dome biogas digester

Recovering energy from biomass and treating organic waste can be done effectively through a biogas digester system. However, monitoring the performance and control of the system to produce biogas is possible through the online process. Therefore, this study developed a novel low-cost gas and temperature profiling system (GTPS) and data acquisition system (DAS) used in biogas digester. The GTPS is designed to monitor and record the amount of methane (CH4) and carbon dioxide (CO2), the main constituent of biogas, from 0 to 100% and to measure the slurry temperature in the digester using a K-type thermocouple. To test the developed monitoring system, the biogas yield (%) and temperature were monitored, and the result obtained agreed with that reported and obtained in the literature (50–70% and 30−40%) for CH4 and CO2, respectively, followed by the desired mesophilic temperature condition for biogas production (20–35 °C). The study employed an RS-232/RS-485 converter for data communication and a low-cost thermocouple module (isolated analog input) that was installed in the data acquisition system. Interestingly, the application of the converter equipment has never been used in any study on monitoring a biogas digester.

Light Output Function and Pulse-Shape Discrimination Capability of p-Terphenyl Organic Scintillator in Wide Neutron Energy Range of 1.1 to 19 MeV

In this work, we studied the light-output properties, efficiency function, as well as the pulse-shape discrimination (PSD) capability of p-Terphenyl scintillator. The selected solid cylindrical scintillation detector has a thickness of 45 mm and a diameter of 45 mm. Recently presented studies of light-output functions have only been measured for low-neutron energies. Our motivation has been to determine the light output function for p-Terphenyl scintillator more accurately over a wider neutron energy range. The measurements have been carried out with mono-energetic neutron beams in the wide energy range from 1.1 to 19 MeV. The neutron–gamma spectrometric system which we developed has been used for the measurement. The input analog signal from the detector was digitized with a fast 12-bits analog to digital converter with a sampling frequency of 1 GHz. Measured data from the detector are processed into the gamma and neutron spectra. The accurate light output function for the p-Therphenyl scintillator has been calculated. The pulse-shape discrimination capability, as well as the detection efficiency, of a p-Terphenyl scintillator are lower in comparison with a NE-213 equivalent detector.

Enhanced Data Processing and Machine Learning Techniques for Energy Consumption Forecasting

Energy consumption plays a significant role in global warming. In order to achieve carbon neutrality and enhance energy efficiency through a stable energy supply, it is necessary to pursue the development of innovative architectures designed to optimize and analyze time series data. Therefore, this study presents a new architecture that highlights the critical role of preprocessing in improving predictive performance and demonstrates its scalability across various energy domains. The architecture, which discerns patterns indicative of time series characteristics, is founded on three core components: data preparation, process optimization methods, and prediction. The core of this architecture is the identification of patterns within the time series and the determination of optimal data processing techniques, with a strong emphasis on preprocessing methods. The experimental results for heat energy demonstrate the potential for data optimization to achieve performance gains, thereby confirming the critical role of preprocessing. This study also confirms that the proposed architecture consistently enhances predictive outcomes, irrespective of the model employed, through the evaluation of five distinct prediction models. Moreover, experiments extending to electric energy validate the architecture’s scalability and efficacy in predicting various energy types using analogous input variables. Furthermore, this research employs explainable artificial intelligence to elucidate the determinants influencing energy prediction, thereby contributing to the management of low-carbon energy supply and demand.

Laboratory systems for detonation research

Decarbonizing global energy systems is among the most pertinent challenges in the climate crisis. Tightly tied to the economic prosperity of a nation and responsible for more than 75% of global emissions, decarbonization of the energy sector will require well-developed solutions. The emergence of detonation engines can reduce emissions in aviation and gas-fired power generation, with theoretical efficiencies 40% lower than conventional engines. These efficiencies have yet to be realized in practice, due to our lack of understanding of detonation propagation and its instabilities. This project aimed to develop the laboratory systems required to study detonations. To this end, a long rectangular channel was manufactured and outfitted with a gas handling system to inject a mixture, an ignition system, pressure transducers to capture the propagation characteristics, and an optical system enabling Schlieren photography. The gas handling system uses sonic nozzles, thus flow rate is a linear function of upstream line pressure. Each line contains a pressure transducer, microcontroller to process signals, LCD for pressure readout, and a regulator for adjustment of the line pressure. The ignition system uses an ignition box that steps up a 12 V supply to 400 V, an ignition coil raising the voltage to 15-40 kV, and a spark plug to ignite the mixture. Ignition, pressure transducer signal acquisition, and Schlieren photography are all synchronized using a NI-6356 DAQ with LabVIEW. The DAQs analog input ports receive readings from the pressure transducers along the channel, the timing of which and known distance between transducers allows for velocity calculations. The schlieren system utilizes the digital output channels to synchronize camera shutter with an LED driver, and parabolic mirrors, positioned to optimize schlieren imaging. The LED driver requires greater current than produced by the DAQ thus a current amplifier circuit is used to bridge the DAQ and driver.

High-precision high-speed nanopore ping-pong control system based on field programmable gate array.

"Molecular ping-pong," emerging as a control strategy in solid-state nanopore technology, presents a highly promising approach for repetitive measurements of single biomolecules, such as DNA. This paper introduces a high-precision, high-speed nanopore molecular ping-pong control system consisting of a home-built trans-impedance amplifier (TIA), a control system based on a Field Programmable Gate Array (FPGA), and a LabVIEW program operating on the host personal computer. Through feedback compensation and post-stage boosting, the TIA achieves a high bandwidth of about 200kHz with a gain of 100 MΩ, along with low input-referred current noise of 1.6 × 10-4pA2/Hz at 1kHz and 1.1 × 10-3pA2/Hz at 100kHz. The FPGA-based control system demonstrates a minimum overall response time (tdelay) of 6.5 μs from the analog input current signal trigger to the subsequent reversal of the analog output drive voltage signal, with a control precision of 1 μs. Additionally, a LabVIEW program has been developed to facilitate rapid data exchange and communication with the FPGA program, enabling real-time signal monitoring, parameter adjustment, and data storage. Successful recapture of individual DNA molecules at various tdelay, resulting in an improvement in capture rate by up to 2 orders of magnitude, has been demonstrated. With unprecedented control precision and capture efficiency, this system provides robust technical support and opens novel research avenues for nanopore single-molecule sensing and manipulation.

Design and Implementation of a Low-Power Device for Non-Invasive Blood Glucose

Glucose is a simple sugar molecule. The chemical formula of this sugar molecule is C6H12O6. This means that the glucose molecule contains six carbon atoms (C), twelve hydrogen atoms (H), and six oxygen atoms (O). In human blood, the molecule glucose circulates as blood sugar. Normally, after eating or drinking, our bodies break down the sugars in food and use them to obtain energy for our cells. To execute this process, our pancreas produces insulin. Insulin “pulls” sugar from the blood and puts it into the cells for use. If someone has diabetes, their pancreas cannot produce enough insulin. As a result, the level of glucose in their blood rises. This can lead to many potential complications, including blindness, disease, nerve damage, amputation, stroke, heart attack, damage to blood vessels, etc. In this study, a non-invasive and therefore easily usable method for monitoring blood glucose was developed. With the experiment carried out, it was possible to measure glucose levels continuously, thus eliminating the disadvantages of invasive systems. Near-IR sensors (optical sensors) were used to estimate the concentration of glucose in blood; these sensors have a wavelength of 940 nm. The sensor was placed on a small black parallelepiped-shaped box on the tip of the finger and the output of the optical sensor was then connected to a microcontroller at the analogue input. Another sensor used, but only to provide more medical information, was the heartbeat sensor, inserted into an armband (along with the microprocessor). After processing and linear regression analysis, the glucose level was predicted, and data were sent via the Bluetooth network to a developed APP. The results of the implemented device were compared with available invasive methods (commercial products). The hardware consisted of a microcontroller, a near-IR optical sensor, a heartbeat sensor, and a Bluetooth module. Another objective of this experiment using low-cost and low-power hardware was to not carry out complex processing of data from the sensors. Our practical laboratory experiment resulted in an error of 2.86 per cent when compared to a commercial product, with a hardware cost of EUR 8 and a consumption of 50 mA.

Laser-sound reproduction by pulse amplitude modulation audio streams

Recently, the possibility to reproduce complex continuous acoustic signals via pulsed laser-plasma sound sources was demonstrated. This was achieved by optoacoustic transduction of dense laser pulse trains, modulated via single- or multi-bit Sigma–Delta, in the air or on solid targets. In this work, we extend the laser-sound concept to amplitude modulation techniques. Particularly, we demonstrate the possibility of transcoding audio streams directly into acoustic pulse streams by analog pulsed amplitude modulation. For this purpose, an electro-optic modulator is used to achieve pulse-to-pulse amplitude modulation of the laser radiation, similarly to the multi-level Sigma–Delta method. The modulator is directly driven by the analog input stream through an audio interface. The performance of the system is evaluated at a proof-of-principle level for the reproduction of test audio signals such as single tones, double tones and sine sweeps, within a limited frequency range of the audible spectrum. The results are supported by computational simulations of the reproduced acoustic signals using a linear convolution model that takes as input the audio signal and the laser-generated acoustic pulse profile. The study shows that amplitude modulation allows for significant relaxation of the laser repetition rate requirements compared to the Sigma–Delta-based implementation, albeit at the potential cost of increased distortion of the reproduced sound signal. The nature of the distortions is analyzed and a preliminary experimental and computational investigation for their suppression is presented.

A frequency meter based on Costas FLL for resonant accelerometers

The resonant accelerometer (RA) is an important part for high-performance inertial devices. And high accuracy frequency measurement is one of the key technology for RA. This paper developed a frequency meter based on Costas frequency lock loop (FLL). This structure is flexible as the analog input sinusodial voltage wave is converted into digital domain by an ADC and the frequency measurement function dealing with that voltage could be implemented only with digital circuits. The transfer function of the frequency meter is determined by the loop filter of the FLL and the closed system is always stable for PID controller. The noise in the loop is reshaped by a first order ∑-Δ modulator formed by the frequency meter loop, which relaxes the requirement for the accuracy of ADC. A test board for the frequency meter is implemented with a 24-bit ADC and FPGA, and achieves a noise of 1mHz/Hz at low frequency band, 7µHz/Hz around 50 Hz with a 40 kHz sinusodial input voltage.

LiKey: Location-independent keystroke recognition on numeric keypads using WiFi signal

As an application of wireless sensing, keystroke recognition can be used in a variety of scenarios, such as password protection and analog input for devices. Existing WiFi-based keystroke recognition systems have strong dependencies on environment and location. Once the environment or keyboard position changes, systems need to re-collect data and re-train the model, otherwise the performance will degrade significantly.In this paper, we propose LiKey, a location-independent keystroke recognition system for numeric keypads, consisting of one transmitter and two receivers. We specially place the transceiver devices, which makes it easy to analyze the variation of the reflection path length for keystrokes at different locations. The same keystroke at different positions has the same change rule of reflection path length, corresponding to the same change rule of rotational direction in the channel state information (CSI) complex plane. We design a method named SRDF to sum the rotational direction features in the CSI complex plane and obtain the location independent features. Since different keystrokes behave differently on the two receivers, a dual-receiver phase–amplitude joint optimization segmentation algorithm is designed to extract the complete keystroke segments. Finally, the dual-branch CNN model is used for training and classification based on the features of the two receivers. The experimental results show that LiKey has strong robustness to the keystroke position. It recognizes well even for keystrokes on positions not involved in the training.

PLC BASED SCADA SYSTEM DESIGN FOR INSTANT MONITORING AND EARLY WARNING MECHANISM OF TOXIC GASES IN UNDERGROUND MINES

In this study, a real-time toxic gas measurement and early warning mechanism is proposed to prevent accidents that may occur in underground mining facilities. Temperature, oxygen (O2), hydrogen sulfide (H2S), methane (CH4), carbon monoxide (CO) and dust density sensor data are read via Arduino Mega analog inputs. All sensor data is transferred to the PLC-SCADA (Programmable Logic Controller - Supervisory Control and Data Acquisition) system via ethernet in real time. Thanks to the designed SCADA screen, sensor values can be monitored instantly. In addition, it can be checked whether the system is alarm and whether the ventilation system works. According to the underground coal mine regulations, work cannot be done in places where the mine air contains less than 19% oxygen, more than 2% methane, more than 50 ppm (0.005%) carbon monoxide and other hazardous gases. The highest permissible hydrogen sulfur ratio for 8 hours of operation is 20 ppm (0.002%). An early warning mechanism was created with the algorithm written taking these limits into consideration. The tests of the designed system were carried out in a laboratory environment and successful results were observed.

A 96 dB DR Second-Order CIFF Delta-Sigma Modulator with Rail-to-Rail Input Voltage Range

A second-order delta-sigma modulator (DSM) is proposed for readout integrated circuits of sensor applications requiring a small area and low-power consumption. The proposed second-order CIFF DSM with the architecture of cascaded-of-integrator feedforward (CIFF) basically consists of two integrators, a 3-bit quantizer, data-weighted averaging (DWA) circuit, and clock generator. The use of the 3-bit quantizer instead of the single-bit quantizer reduces the size of the feedback capacitor in the first integrator. The 3-bit quantizer is designed based on a successive approximation register analog-to-digital converter for small area and low power implementation. Furthermore, the proposed second-order CIFF DSM has a single supply without an additional reference driver while having a wide analog input voltage range with rail to rail. The proposed second-order CIFF DSM, implemented using a 130 nm 1-poly 6-metal CMOS process with a supply of 1.5 V, has an area of 0.096 mm2. It has a sampling frequency of 500 kHz for the implementation of an input bandwidth of 2 kHz and an oversampling ratio of 125. The measured peak signal-to-noise and distortion ratio is approximately 90 dB when the differential analog input signal has a frequency of 353 Hz and an amplitude of 1.2 Vpp. The measured dynamic range is approximately 96.3 dB.

Python Red Pitaya Lockbox (PyRPL): An open source software package for digital feedback control in quantum optics experiments.

We present the Python Red Pitaya Lockbox (PyRPL), an open source software package that allows the implementation of automatic digital feedback controllers for quantum optics experiments on commercially available, affordable Field-Programmable Gate Array (FPGA) boards. Our software implements the digital generation of various types of error signals, from an analog input through the application of loop filters of high complexity and real-time gain adjustment for multiple analog output signals, including different algorithms for resonance search, lock acquisition sequences, and in-loop gain optimization. Furthermore, all necessary diagnostic instruments, such as an oscilloscope, a network analyzer, and a spectrum analyzer, are integrated into our software. Apart from providing a quickly scalable, automatic feedback controller, the lock performance that can be achieved by using PyRPL with imperfect equipment, such as piezoelectric transducers and noisy amplifiers, is better than the one achievable with standard analog controllers due to the higher complexity of implementable filters and possibilities of nonlinear operations in the FPGA. This drastically reduces the cost of added complexity when introducing additional feedback loops to an experiment. The open-source character also distinguishes PyRPL from commercial solutions, as it allows users to customize functionalities at various levels, ranging from the easy integration of PyRPL-based feedback controllers into existing setups to the modification of the FPGA functionality. A community of developers provides fast and efficient implementation and testing of software modifications.

Design and Implementation of Sensored Brushless DC Motor Control Using dsPIC30F4012 for CW/CCW Bidirectional Rotation

This paper presents the development of BLDC (Brushless DC) motor control based on dsPIC30F4012. The system is designed to control motor rotation in clockwise (CW) and counter clockwise (CCW) directions. There are 3 input pins used to facilitate the control interface, namely ground (GND) pin, control (CTRL) pin and enable (EN) pin. An analog input was assigned as command input for rotation in CW/CCW direction. The CTRL pin input has full-scale of analog input that comprises of three regions, namely CCW rotation input (0V~2.4V), CW rotation input (2.6V~5V), and neutral input (2.5±0.1)V as the deadband. The commutation sequences of six steps for direction of motor rotation, while P-I (Proportional-Integral) controller is used to control the motor current. The control law and commutation sequences are implemented in single chip dsPIC30F4012 microcontroller. To verify the system performance, we test the system implementation to drive BLDC motor and measured current are sent to the host PC through serial communication. From the experimental results it is shown that the realizing P-I controller is good performance in small frequency where the bldc motor current can follow the input commands. The bandwidth of motor control to characterize the performance of designed system is also presented.

Development of embedded fuzzy control using reconfigurable FPGA technology

The purpose of this work is to investigate the construction of fuzzy embedded control systems combining fast execution and parallel processing capabilities provided by Field Programmable Gate Arrays (FPGAs) and Reconfigurable Inputs/Outputs (RIO) chips. A fixed-point fuzzy controller is developed and implemented on a fast mechatronic system with high-speed control and high channel count on an FPGA target. This paper provides a brief introduction to deploying Fuzzy Logic Control (FLC) methods using RIO-FPGA technology. It suggests a technique for implementing the three stages that constitute the FLC and the PD-like FLC and PID-like FLC structures into practice. Controllers with 1 and 2 degrees of freedom are developed and tested experimentally. Parallel loops, key challenging advantage of LabVIEW programming, are utilized for decoding feedback signals, generating pulse trains for actuator's drivers and for calculation of control gains. An NI-SbRIO board that combines deployable devices with a real-time processor, a re-configurable FPGA, and analogue and digital input–output ports is used. The experimental work demonstrates the significant enhancement of implementing reconfigurable embedded fuzzy control upon such mechatronic systems.

A High ENOB 14-Bit ADC without Calibration

This paper presents an implementation of a 14-bit 2.5 MS/s differential Successive-Approximation-Register (SAR) analog-to-digital converter (ADC) to be used for sensing multiple analog input signals. A differential binary-weighted with split capacitance charge-redistribution capacitive digital-to-analog converter (CDAC) utilizing the conventional switching technique is designed, without using any calibration mechanism for fast power-on operation. The CDAC capacitor unit has been optimized for improved linearity without calibration technique. The SAR ADC has a differential input range 3.6 Vpp, with a SNDR of 80.45 dB, ENOB of 13.07, SFDR of 87.16 dB and dissipates an average power of 0.8 mW, while operating at 2.5 V/1 V for analog/digital power supply. The INL and DNL is +0.22/−0.34 LSB and +0.42/−0.3 LSB, respectively. A prototype ADC has been fabricated in a conventional CMOS 65 nm technology process.

2Memristor-1Capacitor Integrated Temporal Kernel for High-Dimensional Data Mapping.

Compact but precise feature-extracting ability is core to processing complex computational tasks in neuromorphic hardware. Physical reservoir computing (RC) offers a robust framework to map temporal data into a high-dimensional space using the time dynamics of a material system, such as a volatile memristor. However, conventional physical RC systems have limited dynamics for the given material properties, restricting the methods to increase their dimensionality. This study proposes an integrated temporal kernel composed of a 2-memristor and 1-capacitor (2M1C) using a W/HfO2/TiN memristor and TiN/ZrO2/Al2O3/ZrO2/TiN capacitor to achieve higher dimensionality and tunable dynamics. The kernel elements are carefully designed and fabricated into an integrated array, of which performances are evaluated under diverse conditions. By optimizing the time dynamics of the 2M1C kernel, each memristor simultaneously extracts complementary information from input signals. The MNIST benchmark digit classification task achieves a high accuracy of 94.3% with a (196×10) single-layer network. Analog input mapping ability is tested with a Mackey-Glass time series prediction, and the system records a normalized root mean square error of 0.04 with a 20×1 readout network, the smallest readout network ever used for Mackey-Glass prediction in RC. These performances demonstrate its high potential for efficient temporal data analysis.

A 0.55-mm2 8-bit 32-GS/s TI-SAR ADC with optimized hierarchical sampling architecture

This paper analyzes the bandwidth of the time-interleaved analog-to-digital converter (TI-ADC) with hierarchical sampling and presents an 8-bit 32-GS/s TI-ADC in 28-nm CMOS. The front-end sampler is implemented by a two-stage hierarchical sampling architecture to extend the analog input bandwidth. The 32-way sub-ADCs are realized with 2b/cycle successive approximation register (SAR) architecture and non-binary search strategy, which meet the requirements of high speed, medium resolution, and strong robustness. Multi-phase low-jitter clock tree is designed based on the delay-locked loop (DLL). Calibration methods for timing skew, offset and gain mismatches are introduced to improve accuracy. The proposed TI-SAR ADC achieves an analog input bandwidth of 23.7 GHz. It occupies an active area of 0.55 mm2 and consumes 356 mW at 1.8/1 V supply.

MEASURING THE SMALL LINEAR DISPLACEMENTS BY THE PHASE METHOD

A measuring device with a differential capacitive measuring transducer (CMT) designed to measure small linear displacements is considered. The CMT has a plane-parallel design of the plates, the natural input value is the change in the gap between the plates. In the measuring circuit (MC), model resistors with the same resistance are connected in series with the sensitive elements of the CMT. The voltage dividers formed by these connections are powered by counter currents of a high-frequency sinusoidal voltage generator. An informative measurement parameter is the phase shift angle between the two output voltages of the MC. These voltages are supplied to the analog inputs of the programmable microcontroller (PMC). The PMC controls the measurement process, measures the specified phase shift angle and determines the measured displacement. Thanks to the used algorithm for processing the measurement results, a ratiometric output signal is obtained to determine the measured displacement. This ensures high metrological characteristics of the meter. The theory of the meter's functioning, the features of the choice of nodes, and the method of calculating the parameters of the MC elements are described. A methodology for the theoretical study of measurement errors has been developed, and the results of their calculation for the implemented version of the meter are presented. It is substantiated that the relative measurement error of the developed device for small linear displacements does not exceed 0.065%. Ensuring high measurement accuracy is also due to the use of phase signals instead of potential-current signals.

Diagnostic system of the facility for studying superconducting quadrupoles

For the HED@FAIR project, NRC “Kurchatov Institute” — IHEP creates four quadrupole magnets with a unique combination of a large internal diameter of the superconducting winding and a high magnetic field gradient in the magnet aperture. To evaluate these superconducting magnets in different operation modes (cooling, maintaining operating parameters, heating, and transitioning magnet to the normal state), a cryogenic facility equipped with a diagnostic system was created. The main measured parameters of the unit were temperature, pressure, and vacuum in the equipment and pipelines, levels of liquid helium and liquid nitrogen, level of the gas holder, and helium flow rate. The control devices were represented by valves with electric and valves. For processing incoming signals from sensors and generation of control actions, the equipment of the domestic company OVEN was used, particularly analog input modules MV110, discrete output modules MU110, and interface converters alternating current four and direct current power supplies. The critical part of the diagnostics system, namely, the control of cryogenic valves, was provided through analog output modules NI 9219 and input modules NI 9212 (feedback) of the CompactRIO chassis. The program part of the installation for the study of superconducting quadruples is a mnemonic diagram with interactive graphic elements, physically located on the computer (operator console), and available in the read mode via the remote desktop protocol to users outside the installation. The mnemonic scheme allows the manual control of the plant’s valves and the automatic regulation of some indicators, depending on the settings assigned by the operator. The dynamics of the processes can be monitored with graphs on the right side of the interface. These graphs are displayed on a separate display at the operator’s console, whereas they are displayed on a unified screen for remote users.

(Invited) Physical Understanding of HfZrO2/Si FeFET Memory and Its AI Applications

Since the discovery of ferroelectricity in HfO2-based dielectric films in 2011 [1], FeFETs using HfO2-based ferroelectric films as gate dielectrics have attracted strong interest and have been intensively studied in the worldwide research community for various applications including memory, logic, and AI computing. In this paper, we present our recent activities on HfZrO2 (HZO)/Si FeFETs memory [2-6] and FeFET reservoir computing [7-9] for low-power memory and AI applications.HZO-based FeFETs are currently expected for a variety of applications such as embedded nonvolatile memory, cross-bar Commutation-in-Memory (CiM), and so on. However, the device operation has not been quantitatively understood yet. One of the fundamental questions in FeFETs is a mismatch between areal surface carrier concentration (Ns) in MOSFETs, ~1013 cm-2 at maximum, and typical polarization (P) values, ~10-20 mC/cm2. These polarization values correspond to Ns higher than 1014 cm-2 and induce an electric field across an interfacial layer (IL) of 30 MV/cm or higher, which could cause oxide breakdown. On the other hand, when a large amount of traps is included in an HZO film or an HZO/IL interface, sufficient P values can be obtained without too high Ns and the electric field [2, 4-6]. Thus, it is very important to experimentally characterize P, Ns, and trapped charge density, which were evaluated by P-V, quasi-static split C-V, and Hall measurements, respectively, [2, 3, 6] for 10-nm-thick HZO/Si n- and p-FeFETs with 0.7-nm-thick SiO2 IL. As a result, it has been found that a high trap density of around 1014 cm-2 is observed in n-FeFETs, while there is almost no hole trapping in p-FeFETs. This high electron trap density leads to the high electric field across the HZO film and resulting high P values in n-FeFETs, which is in contrast to p-FETs. Thus, the larger memory window in n-FeFETs than that in p-FeFET is attributed to the difference in P values with applying the positive and negative gate voltages.On the other hand, we have proposed and demonstrated reservoir computing, which is one of AI calculations using a recurrent neural network, by using HZO/Si FeFETs [7-9]. Reservoir computing has recently been attracting strong attention as a method to realize real-time learning of time-series data at the edge. Furthermore, when this reservoir is represented by hardware that has input-history-dependent and nonlinear dynamics (physical reservoir), we can expect significant improvement in the energy efficiency of AI calculations. In our FeFET reservoir, the time response of a drain current of Si FeFETs was utilized as the virtual nodes. We have experimentally demonstrated high classification accuracy in speech recognition by using this HZO/Si FeFET reservoir computing with parallel data processing [9]. The spoken digit recognition task, which is an example of speech recognition applications, was performed on a basis of multiple frequency channel (cochleagram) data, which were input to multiple FeFETs. The final inference was conducted by a majority vote of the multiple FeFET reservoirs. We introduced several approaches in the proposed FeFET reservoir computing system, such as optimization of the number of virtual nodes per time step, analog and digital input waveforms, the utilization of other currents than the drain current, and the combination of different frequency channels, in order to improve the classification accuracy. The choice of a combination of frequency channels was effective in improving recognition accuracy. Also, combining the time responses of the drain, source, and substrate currents was also important. These various methods additively improved the recognition accuracy. As a result, we have experimentally demonstrated a classification accuracy of 95.9 % in the task to classify the audio waveforms of ‘0’ to ‘9’ spoken digits.This work was supported by JST CREST (JPMJCR20C3) and JSPS KAKENHI (21H01359).