LAB51 logo

IBM Research Unveils Revolutionary AI Chip for Deep Learning

IBM Research analog AI chip
By Nicolo Finazzi
Nicolo Finazzi

2 Min

August 14, 2023
  • IBM Research introduces analog AI chip for efficient deep neural network computations.
  • Chip leverages nanoscale resistive memory devices to emulate neural network functions.
  • 92.81% accuracy achieved on CIFAR-10 dataset, showcasing the chip's power.

IBM Research has unveiled a revolutionary analog AI chip that represents a significant advancement in the realm of high-performance artificial intelligence computing. The chip marks a significant leap forward in achieving energy-efficient AI computations without compromising computational accuracy.

Traditional digital computing architectures face limitations when executing deep neural networks (DNNs) due to the constant data transfer between memory and processing units. This impedes performance and energy optimization. To overcome these challenges, IBM Research has embraced the principles of analog AI, mirroring the neural network operations found in biological brains.

By harnessing nanoscale resistive memory devices, specifically Phase-change memory (PCM), the chip emulates synapse behavior in biological neural networks. PCM devices modify their conductance through electrical pulses, offering a continuum of values for synaptic weights. This analog approach eliminates the need for extensive data transfers, enhancing computational efficiency.

The chip's performance was demonstrated with remarkable accuracy, achieving an unprecedented 92.81 percent on the CIFAR-10 image dataset. Moreover, its compute efficiency, measured in Giga-operations per second (GOPS) per unit area, outperformed previous in-memory computing chips.

IBM Research's breakthrough in analog AI presents a promising pathway toward energy-efficient AI computing across various applications, setting a significant milestone in AI hardware development.

To know more about these topics, read the following articles: