School of Engineering Department of Electronic and Computer Engineering 174 Design and Characterization of a Tactile Sensor Array Supervisor: WONG Man / ECE Student: GUO Shiran / CPEG LI Liwei / CPEG ZHOU Hangan / CPEG Course: UROP 1100, Fall This report presents the design and characterization of a tactile sensor array, an electromyography system, and their data processing methods. The tactile sensor combines polyvinylidene fluoride with dual-gate indium-tin-zinc oxide thin-film transistors (TFT) to create a high-gain amplifier. An experimental setup simulates a robotic hand interacting with Braille, with data collected via NI DAQ and analyzed using LabView software. The EMG system captures the activity from the lower arm to allow indirect hand motion estimation. Data processing is performed with a compute-in-memory (CIM) synapse circuit using TFTs and capacitors for data storage and retrieving during ANN computation. Design and Characterization of a Tactile Sensor Array Supervisor: WONG Man / ECE Student: GUO Shiran / CPEG Course: UROP 2100, Spring To verify the computational power of thin-film transistors (TFTs) and their potential for flexible device fabrication, we performed ANN computation using a single-layer perceptron on a 32x32 TFT circuit array based on organic flexible materials to classify electrocardiogram (ECG) signals. The first part of the report illustrates the procedure for hardware testing, algorithm design and implementation, and analysis of the results. Spiking Neural Network (SNN) is a neuromorphic computing algorithm. In order to evaluate the similarity between the algorithm and the performance of the human brain, we designed an experiment to collect data for the performance of humans in real-time recognition of handwritten digits for human-machine comparisons. The methodology and results are presented in the second part. Design and Characterization of a Tactile Sensor Array Supervisor: WONG Man / ECE Student: LI Liwei / CPEG Course: UROP 2100, Spring This study optimizes a third-generation Spiking Neural Network (SNN) for real-time handwritten digit prediction through software simulation, modeling Thin-Film Transistor (TFT) circuits. Eight parameters, including STDP_A and Leakage_TC, were tuned using a systematic method involving coarse and fine adjustments, achieving a training accuracy of 91.5% and a test accuracy of 92.3%. Novel metrics, Test_pre and Train_pre, were introduced to evaluate early prediction during the writing process, yielding values of 0.669 and 0.662, respectively. A second experiment with raw data showed reduced accuracy (66.57% training, 63.67% test), highlighting the importance of feature extraction. These results demonstrate SNNs’ potential for efficient, biologically inspired computing, with implications for future TFT-based hardware.
RkJQdWJsaXNoZXIy NDk5Njg=