Task 003: Algorithms for Emerging Hardware
|Event Date:||February 13, 2020|
|School or Program:||Electrical and Computer Engineering
"Designing an Approximate Memristive In-Memory Hamming Distance Circuit for Low-power Inference"
Abstract: Hamming Distance (HD) is a popular similarity measure that is used widely in binary-pattern matching applications, DNA sequencing, and binary error-correcting codes. Current HD circuits suffer from high latency and/or high power consumption due to having separate memory and computing modules, and/or using power-hungry CMOS circuitry. In this talk, I will present our memristive in-memory approximate HD circuit that uses the memristive crossbar’s reverse/sneak currents to compute the HD. The HD computation accuracy of our circuit, under non-ideal fabrication conditions, is 97% compared to other HD circuits’ 100%; while consuming on average ≈ 300x less power. Our circuit’s operation is independent of the memristor model used, as long as the model allows a reverse current. Because we leverage in-memory parallel computing, our circuit is n times faster than other HD circuits, where n is the number of HDs to be computed. Our low-power and fast HD circuit is relevant for binary-pattern matching and DNA sequencing applications.
Bio: Mohammad is currently a PhD candidate at Portland State University, working with Prof. Teuscher. He received his MSc in Electrical Electronics Engineering from University of Greenwich, UK. His research interests include approximate and probabilistic computing, machine learning, and low power computing architectures.