NeuroBlade Preps Inference Chip

Article By : Rick Merritt

NeuroBlade emerged from stealth mode with plans to sample an AI inference accelerator by early 2020, with additional plans for a follow-on training chip.

SAN JOSE, Calif. — Add NeuroBlade to the dozens of startups working on AI silicon. The Israeli company just closed a $23 million Series A, led by the founder of Check Point Software and with participation from Intel Capital.

NeuroBlade is close to taping out an inference processor that it aims to sample late this year or early in 2020. Following in the footsteps of Israeli AI startup Habana, it plans a training accelerator using a similar architecture for its next generation.

The startup claims that its initial chip will run several neural networks and multiple algorithms in parallel. The chip supports a full range of neural-net models with fully connected layers, including CNNs, RNNs, LSTMs, and GANs.

The chip will provide similar performance on inference jobs as Intel CPUs or Nvidia GPUs at a lower cost thanks to a much smaller die, according to the company. Alternatively, it will support significantly improved performance with a similar size and cost die.

NeuroBlade declined to provide any public benchmark results until chips are shipping. However, it said that it is sharing performance metrics with select customers.

Founders Elad Sity, CEO, and Eliad Hillel, CTO, met while working at Solar Edge, a startup that went public making inverters and other gear for photovoltaic arrays. Both are graduates of the technology unit of the Israel Intelligence Corps.

NeuroBlade got its initial start with a $4.5 million seed round in 2017 from StageOne Ventures and Grove Ventures, headed by Dov Moran, the inventor of the USB flash drive and co-founder of M-Systems.

Related posts: