Neuromorphic Computing-in-Memory Chip NeuroCore-X Launched: AI Inference Energy Consumption Reduced to One-Twentieth of Traditional GPUs

On August 9, 2028, Tsinghua University's Center for Brain-Inspired Computing released the NeuroCore-X neuromorphic chip at the International Solid-State Circuits Conference (ISSCC).

NeuroCore-X uses computing-in-memory (CIM) architecture, embedding compute units directly within storage arrays to avoid the "memory wall" bottleneck caused by repeatedly shuffling data between processor and memory in traditional architectures. The chip also uses spiking neural networks (SNN) rather than traditional artificial neural networks, generating spike signals only when needed to further reduce power consumption.

On ImageNet inference benchmarks, NeuroCore-X's performance-per-watt was 20x that of the NVIDIA H100 GPU, with absolute performance approximately 3x the H100. The chip has been successfully taped out, with small-batch production expected in Q1 2029.

Team lead Professor Luping Shi said NeuroCore-X is particularly suited for edge AI inference scenarios, such as autonomous vehicle computing, smartphone AI assistants, and industrial quality inspection equipment.