Stimulus Truncation Method for Energy Efficient Memristor Based Neuromorphic Computing
TimeTuesday, December 7th6:00pm - 7:00pm PST Event Type
DescriptionA solution is presented for improving energy efficiency and heightening the performance of neuromorphic computing systems. Specifically, a pulse truncation (PT) method is proposed to reduce number of pulses in every weight update. The neuromorphic computing system with the PT method is implemented and evaluated based on the fabricated memristor and its tested current-pulse characteristics. The results show that the PT method cannot only alleviate uneven pulse distribution in writing procedure, but also save the writing energy of crossbar array by 8.29%-26.87% and reduce the writing latency by 30%-48%. Finally, the significant nonlinearity and many variations are considered.