One of the most appealing characteristics of Phase-Change Memory (PCM) technology is the ability to store analog quantities as conductances. To fruitfully exploit this feature in Analog in-Memory Computing (AiMC) applications, highprecision programming algorithms are required. In this paper, a method to reach a fairly arbitrary number of analog levels in PCM cells is proposed. The algorithm has been validated on an AiMC prototype, designed and fabricated using a 28-nm FD-SOI process by STMicroelectronics to compute 512 × 512 signed Matrix-Vector Multiplications (MVMs), achieving an Equivalent Number of Bits (ENOB) of 10.55, which improves state-of-theart weights programming accuracy for PCM-based hardware accelerators.
High-Precision Close-to-Analog Programming of PCM Cells as Devices for AiMC Edge-AI
Zurla, R.;Vignali, R.;Iannelli, L.;Cabrini, A.;
2025-01-01
Abstract
One of the most appealing characteristics of Phase-Change Memory (PCM) technology is the ability to store analog quantities as conductances. To fruitfully exploit this feature in Analog in-Memory Computing (AiMC) applications, highprecision programming algorithms are required. In this paper, a method to reach a fairly arbitrary number of analog levels in PCM cells is proposed. The algorithm has been validated on an AiMC prototype, designed and fabricated using a 28-nm FD-SOI process by STMicroelectronics to compute 512 × 512 signed Matrix-Vector Multiplications (MVMs), achieving an Equivalent Number of Bits (ENOB) of 10.55, which improves state-of-theart weights programming accuracy for PCM-based hardware accelerators.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
