The startup's neuromophic processing unit combines a training algorithm with a new analog AI chip architecture.
Rain Neuromorphics, the startup working on a full-blown analog AI chip, has raised $25 million in a Series A funding round. The company plans to invest the funds in product development while tripling its engineering and support staff.
The round was led by Prosperity 7 Ventures along with existing investors Buckley Ventures, Gaingels, Loup Ventures, Metaplanet and Pioneer Fund, among others. Backers also include angel investors Sam Altman, co-founder and CEO of OpenAI, and Jeff Rothschild, founding engineer of Facebook. Altman previously led Rain Neuromorphic’s seed round in 2018.
The startups’ neuromorphic processing unit (NPU) combines a new training algorithm – Equilibrium Propagation – with a new analog chip architecture. The combination can accelerate processing and reduce power consumption, with an overall energy reduction by a factor of 1,000 compared to current AI systems, the company claims. While analog computing is used today in some processor-in-memory chips, the approach requires energy-hungry ADCs and DACs between network layers. Analog computing in its current form is also incompatible with back-propagation, the algorithm widely used for training.
Rain’s NPU uses resistive RAM (ReRAM) as a memristive element, then combines it with 3D manufacturing techniques and vertical bit lines borrowed from NAND flash technology. The approach allows Rain to create a chip modeled on the structure of brain cells. The vertical bit lines, coated in a memristive material, are analogous to axons, CMOS layers below represent neurons and randomly configured sparse connections between axons and neurons are analogous to dendrites.
The point where a dendrite in any layer contacts a column can be thought of as a synapse.
An NPU test chip taped out earlier this year is built on a 180-nm CMOS process with 10,000 neurons. It has already demonstrated training and inference capabilities.
Rain Neuromorphic’s long-term goal is creating a brain-like chip that can initially be used in both cloud and edge AI applications.
This article was originally published on EE Times.
Sally Ward-Foxton covers AI technology and related issues for EETimes.com and all aspects of the European industry for EE Times Europe magazine. Sally has spent more than 15 years writing about the electronics industry from London, UK. She has written for Electronic Design, ECN, Electronic Specifier: Design, Components in Electronics, and many more. She holds a Masters’ degree in Electrical and Electronic Engineering from the University of Cambridge.