A Comparative Analysis of Aging Effects on Computing-in-Memory Systems in 28nm CMOS

This paper presents a 28nm CMOS simulation study that explores aging-induced accuracy and frequency degradation in analog Compute-In-Memory (ACIM) and digital Compute-In-Memory (DCIM) systems for machine learning accelerators. We show that Bias Temperature Instability (BTI) has a greater impact than Hot Carrier Injection (HCI). Simulation results also show that Analog-Input-Analog-Computing (AIAC) is more vulnerable to aging than Digital-Input-Analog-Computing (DIAC) and DCIM.