NeuDW-CIM: a 65-nm 0.8-pJ/Sop Reconfigurable Neuromorphic Compute-in-Memory Macro with Nonlinear Dendrites and K-Winners
2026-06-08 • Hardware Architecture
Hardware Architecture
AI summaryⓘ
The authors created NeuDW-CIM, a special computer chip designed to efficiently run Spiking Neural Networks, which mimic how the brain works. They used a new type of memory cell that can handle three states instead of two and a smart converter inside the memory. Their chip works in two modes: one mimics brain-like processing with high accuracy, and the other speeds up computation by stopping early when enough information is found, saving energy. Overall, their design makes running these brain-inspired networks faster and more energy-efficient.
Neuromorphic ComputingCompute-in-MemorySpiking Neural NetworksTernary Bit-cellIn-Memory ADCNonlinear DendriteTop-K WinnerEnergy EfficiencyCMOS TechnologyLatency
Authors
Junyi Yang, Yahan Yang, Shuai Dong, Biyan Zhou, Ye Ke, Zhengnan Fu, Xin Si, An Guo, Peng Zhou, Arindam Basu
Abstract
This work presents NeuDW-CIM, a highly efficient neuromorphic Compute-in-Memory (CIM) macro for Spiking Neural Networks (SNNs) implemented in 65 nm CMOS. The design introduces a custom twin 9T bit-cell for ternary in-puts/weights and a reconfigurable non-linear In-Memory ADC (IMA). The macro supports two specialized modes: 1) Nonlinear Dendrite (NLD) mode, which utilizes reconfigurable IMA to emulate biological dendritic functions, achieving measured accuracies of 97.2% on N-MNIST and 95.5% on DVS Gesture; and 2) Top-K Winner (KWN) mode, featuring an early-stopping mechanism that reduces IMA conversion latency by 30% and digital LIF latency by 10x. Benefiting from the sparse update in KWN mode, NeuDW-CIM achieves a measured energy efficiency (EE) of 0.8 pJ/SOP (1.6x improvement).