By Max A. Cherney
SAN FRANCISCO (Reuters) – Chip startup Retym has raised $75 million this year as part of $180 million in total funding the company is using to pursue networking chips for artificial intelligence computing in data centers, the company said on Monday.
Retym makes chips that perform digital signal processing (DSP) that helps move information speedily between large data centers, a capability that has become increasingly important with the surge of interest in AI.
Creating the underlying AI models that power the likes of ChatGPT requires thousands of chips to be strung together with networking equipment.
At the moment the market for the DSP chips Retym is working on is dominated by Marvell Technology.
The chip Retym – pronounced re-time – is working on will help solve a bottleneck that has developed in data centers, chief executive Sachin Gandhi said. Because large numbers of chips must work together on AI computing tasks the connectivity between them is increasingly important.
“We are focusing on building coherent DSP chips for the next generation deployment of AI infrastructure and cloud,” Gandhi said.
Retym’s first chip will be designed to move data around from a range of 10 kilometers and 120 kilometers, but will be optimized for 30 kilometers to 40 kilometers. The DSP chip Retym is building uses a modulation technique to ensure data transported is not corrupted.
“They took the approach of solving the harder problem with longer distances,” said Navin Chaddha, managing partner at Mayfield, a venture fund that has invested in Retym.
Retym is using Taiwan Semiconductor Manufacturing Co.’s five nanometer manufacturing process for the first chip, which its engineers are currently testing and validating samples.
The Series D funding round was led by Spark Capital. Retym launched four years ago and plans to bring its first product to market this year.
(Max A. Cherney in San Francisco. Editing by Gerry Doyle)
Disclaimer: This report is auto generated from the Reuters news service. ThePrint holds no responsibility for its content.