Rivos raises $250M+ to develop chips for AI and analytics workloads

Rivos Inc., a startup developing artificial intelligence chips based on the RISC-V instruction set architecture, has secured more than $250 million in fresh early-stage funding.

The company raised the capital through a Series-A3 round announced today. According to Rivos, Matrix Capital Management was the largest investor. Intel Capital, Dell Technologies Capital and chipmaker MediaTek Inc. contributed as well along with a half-dozen other backers.

Rivos is developing a chip for powering generative AI and analytics applications. According to Bloomberg, the processor will be positioned as a lower-cost alternative to Nvidia Corp.’s market-leading graphics cards. Rivos plans to make its chips using Taiwan Semiconductor Manufacturing Co. Ltd.’s three-nanometer node.

The company has so far shared only high-level information about its processor’s architecture. According to Rivos, it will feature an onboard graphics processing unit known as the Data Parallel Accelerator. The GPU will be supported by central processing unit cores based on the RISC-V instruction set architecture. 

A CPU performs complex tasks such as running AI software by breaking down the workflow into a series of relatively simple computations. Those computations are known as instructions. Some instructions perform a single action, such as adding up two numbers, while others can carry out multiple computing operations.

RISC-V, the technology on which Rivos’ upcoming chip is based, provides a collection of pre-packaged instructions that engineers can incorporate into their CPU designs. It thus removes the need to develop everything from scratch. Hardware teams with advanced requirements can optionally extend RISC-V by adding custom instructions.

The technology is positioned as an alternative to Armv9, a popular instruction set architecture from Arm Holdings plc. Armv9 is proprietary whereas RISC-V is available under an open-source license. As a result, companies such as Rivos that adopt the latter architecture can avoid the licensing fees the British chip designer charges to use its technology.

Both Armv9 and RISC-V implement a computing approach known as RISC, for reduced instruction-set computer. Each instruction run by a RISC processor performs a single, relatively simple computation such as an arithmetic calculation. That contrasts with the traditional approach of having each instruction perform multiple computations. In many cases, RISC’s single-calculation approach facilitates faster processing speeds.

Rivos says its upcoming chip’s GPU and RISC-V CPU will share a “uniform memory across DDR DRAM and HBM.” In some CPUs, each core must create its own copy of the data being processed. Allowing multiple compute modules to keep their information in the same memory pool can reduce the need for data copies, which improves performance.

The company will also provide a set of custom software tools for its chips. It says those tools will reduce the amount of effort involved in deploying AI models on its silicon.

“The rapid changes in LLMs and the merger with the data analytics stack makes it vital that accelerators be easy to program and debug, and that data can seamlessly move between CPU and accelerator,” said co-founder and Chief Executive Officer Puneet Kumar. “Rivos addresses this need through our recompile-not-redesign approach.”

The latest funding round comes two months after it settled a lawsuit with Apple Inc. over its chip technology. The iPhone maker accused Rivos of poaching employees and stealing trade secrets related to handset processors. According to The Information, the litigation complicated the startup’s efforts to raise capital and led it to lay off some employees.

Rivos said that it will use the proceeds from its newly closed funding round to support go-to-market initiatives. The investment will enable the company to finalize the design of its first processor. In parallel, Rivos plans to expand its manufacturing, software engineering and technical support operations.

Photo: Unsplash