Abstract: Most of the energy consumed by artificial intelligence (AI) workloads on conventional computers is spent moving data between the memory and the processor. In-memory computing helps reduce this data movement by storing the synaptic weight of a neural network as device conductance in a crossbar array. The weighted sum of each layer is then obtained in a single read step. Phase-change memory (PCM) is the most matured non-volatile memory technology demonstrating the ability of non-volatile multilevel conductance states for multi-core in-memory computing chips. However, resistance drift and high reset current are the two major challenges that limit the overall performance for multibit storage within the cell.

In this work, we explored novel Te-rich, low-melting, Ge-Te-Si (GTS) based two terminal pore-type PCM devices as an active layer. These devices exhibit five orders of resistance contrast between its fully set low resistance state (LRS) and fully reset high resistance state (HRS). Further, a fixed, open-loop voltage pulse programming scheme is optimized and applied to these GTS memory devices without write-and-verify, at the same compliance current for both partial-set(gradual increase in conductance) and partial-reset (gradual decrease in conductance). This optimized programming scheme produces eight stable resistance levels (3-bit per cell) between ≈ 1 MΩ and ≈ 20 MΩ using only part of the available resistance contrast. These levels show very low drift coefficients of 0.004–0.025 compared to conventional GeSbTe devices and remain distinct over 10⁴ s. Furthermore, data retention tests were performed for 90 days.Event Details

Title: Low drift, multi-level programming in Ge-Te-Si-based phase change synaptic devices for In-memory computing
Date: November 05, 2026 at 04:00 PM
Venue: ESB 210B
Speaker: Ms. Rajeswari Yerraguntla (EE21D015)
Guide: Dr. Anbarasu Manivannan
Type: PHD seminar

Updated: