Breaking

Chinese team unveils memristor chip up to 478x faster than A100 for brain mapping

July 8, 2026 at 00:39 EDT

Infra & Chips
Research & Papers
Healthcare

A research team from Peking University and the Chinese Academy of Sciences has developed a neuromorphic chip based on phase-change memristors, achieving up to 478 times faster processing than an NVIDIA A100 GPU for 3D reconstruction of the brain's cortical surface. The results were published in Science in July 2026.

Brain-Mimicking Chip · Peking University × CAS

A neuromorphic chip reconstructs the human cortex in under half a second

A phase-change memristor "computing-in-memory" chip runs a neural dynamical system entirely inside memory — reporting 50 to 478× the speed of an NVIDIA A100 GPU on specific brain-modeling tasks.

478×

Peak speed vs. A100 GPU on specific tasks

<0.5s

To reconstruct the human cortical surface

2.12ms

Single-step latency — first in the millisecond range

Speed advantage — drawn to scale

Time to complete the same specific task (relative units, A100 = baseline). The memristor chip's column shows its 50–478× range.

1×

NVIDIA A100
GPU baseline

50–478×

Memristor chip
same task, faster

Why it's fast — computing inside memory

Conventional chips shuttle data between separate memory and processor (the von Neumann bottleneck). This design runs the math where the data lives — and turns memristor "conductance drift," usually a defect, into computation.

Noisy / incomplete
brain data

→

In-memory compute
array (0.28 mm²)

→

3D cortical
reconstruction

Process

40 nm

Core array

0.28 mm²

Type

Phase-change memristor

The promise

Early Alzheimer's screening & diagnosis
Brain-computer interfaces
Intraoperative neuronavigation
Patient-specific "digital brain twins"

The caveats

Task-specific, not a general-purpose GPU comparison
Research prototype — not a commercial product
Availability & pricing undisclosed
Clinical validation & scaling still to come

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.