Unicorn-CIM shows that FP16 DNNs on SRAM CIM crash from exponent bit flips, and a shared-exponent fine-tuning plus lightweight Hamming ECC restores resilience with about 9% logic overhead.
A 22nm 832kb hybrid-domain floating-point sram in-memory-compute macro with 16.2-70.2tflops/w for high-accuracy ai-edge devices,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AR 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Unicorn-CIM: Uncovering the Vulnerability and Improving the Resilience of High-Precision Compute-in-Memory
Unicorn-CIM shows that FP16 DNNs on SRAM CIM crash from exponent bit flips, and a shared-exponent fine-tuning plus lightweight Hamming ECC restores resilience with about 9% logic overhead.