A 22nm 16Mb Floating-Point ReRAM Compute-in-Memory Macro with 31.2TFLOPS/W for AI Edge Devices. Abstract: AI-edge devices demand high-precision computation.

Feb 21, 2024 · This work presents a FP 16Mb ReRAM-nvCIM macro utilizing foundry 22nm 1T1R ReRAM devices and achieves 28.7 and 31.2TFLOPS/W using FP16 and BF16 ...

34.8 A 22nm 16Mb Floating-Point ReRAM Compute-in-Memory Macro with 31.2TFLOPS/W for AI Edge Devices. February 2024. DOI:10.1109/ISSCC49657.2024.10454468.

34.8 A 22nm 16Mb Floating-Point ReRAM Compute-in-Memory Macro with 31.2TFLOPS/W for AI Edge Devices. Tai-Hao Wen 1. ,. Hung-Hsi Hsu 1. ,. Win-San Khwa 2.

34.8 A 22nm 16Mb Floating-Point ReRAM Compute-in-Memory Macro with 31.2TFLOPS/W for AI Edge Devices. AIエッジデバイスのための31.2TFLOPS/Wによる34.8A 22nm 16Mb ...

May 29, 2024 · In Paper 34.8, NTHU describes a 16Mb ReRAM-based non-volatile CIM macro in 22nm CMOS, capable of computing in both FP16 and BFloat16 formats. ...

... AI edge processors. Read more. 34.8 A 22nm 16Mb Floating-Point ReRAM Compute-in-Memory Macro with 31.2TFLOPS/W for AI Edge Devices · Conference Paper. February ...

Jun 28, 2024 · ... 34.8 and 30.1. The former is a 22nm ReRAM-CIM macro with three novelties – (1) kernel-wise weight pre-alignment, (2) rescheduled multi-bit ...

An 8-Mb DC-current-free binary-to-8b precision ReRAM nonvolatile computing-in-memory macro using time-space-readout with 1286.4-21.6 TOPS/W for edge-AI devices.

Computing in memory optimizes data handling by performing operations directly in memory, ideal for high-speed data processing needs.