Analog implementation of a Spiking Neural network

הרקע לפרויקט:

In recent years Artificial Neural Networks (ANN) have reached maturity, and AI is becoming a technology used everywhere. ChatGPT , Bard and Bing have vast distribution and ANN farms are growing exponentially. The ANN attempts to mimic the human brain as closely as possible to achieve the best possible computing power. As the brain signals are analog and spikey by nature the next step for Artificial Neural Networks is to implement Analog Spiking Neural networks (A-SNN).The aim of this project is to implement a A-SNN which is competitive with recent state-of-the-art publications.

מטרת הפרויקט:

In this project you will design a highly compact matrix of Analog Spiking Neurons that combined implement a full Analog Spiking Neural networks . The Neuron implementation will include the design of a switched-cap integrator and a comparator with a precise programable threshold (see figure below).Both the switched-cap integrator and the comparator will be based on previous designs. Once the Neuron is designed you will also design an array of Neurons which implements a full Analog Spiking Neural networks.You will need to learn the theory and then implement the circuit in 28nm/0.18u CMOS. The design includes both analog and digital blocks, which will allow you to develop skills in both.

תכולת הפרויקט:

In this project the student will first design a Spiking Neuron and then combine them into a full matrix that becomes a full Analog Spiking Neural Network. The implementation will be done using digital and analog techniques. The schematics will be prepared in Virtuoso and simulated. Layout and post-layout simulations will be conducted to verify the circuit performance.

קורסי קדם:

768330301 אלקטרוניקה לינארית - חובה
768332501 מעבדה למעגלים אנלוגיים – חובה
8330801 מעגלים אלקטרוניים ספרתיים – חובה
83315 מעבדה מעגלים אלקטרוניים ספרתיים – חובה
768361101 מעגלים משולבים אנלוגיים – מומלץ

דרישות נוספות:

This project is a paid research position for students who are interested in working in our lab part time, and also pursuing an MS degree afterwards. This project is together with some industrial partners.

מקורות:

1. Y. Ko, S. Kim, K. Shin, Y. Park, S. Kim, and D. Jeon, “A 65 nm 12.92-nJ/Inference Mixed-Signal Neuromorphic Processor for Image Classification,” IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 70, no. 8, pp. 2804–2808, Aug. 2023
2. M. Ochs, M. Dietl, and R. Brederlow, “An Analog and Time-Discrete Neuron with Charge-Injection for Use in Ultra-Low Power Spiking Neural Networks,” in 2024 19th Conference on Ph.D Research in Microelectronics and Electronics (PRIME). Larnaca, Cyprus: IEEE, Jun. 2024, pp. 1–4
3. J. Song, X. Tang, H. Luo, K. Xu, Y. Wang, Z. Ji, R. Wang,and R. Huang, “Spike-CIM: A 290TOPS/W Spike-Encoding Sparsity Adaptive Computing-in-Memory Macro with Differential Charge-Domain Integrate-and-Fire,” in 2022 IEEE Asian Solid-State Circuits Conference (A-SSCC). Taipei, Taiwan: IEEE, Nov. 2022, pp. 1–3