C-DAC leads India’s push to build an Nvidia-class AI chip

The global artificial intelligence (AI) boom has spurred India to develop its own Nvidia-like chip, with a production-grade version targeted within the next three years, two senior government officials said.

At the heart of this effort is a four-decade-old research institution, for which the chip could mark its most significant contribution yet.

India’s first, indigenous AI inference chip will be made and presented as early as “2029 or 2030″, Union IT minister Ashwini Vaishnaw said on Wednesday after the Cabinet approved a ₹1.27 trillion Semicon 2.0 mission to incentivize chip factories, startups, design firms and research.

While details of the AI inference chip remain sparse, the officials cited earlier said the processor is currently undergoing trial production. For this, the Centre for Development of Advanced Computing (C-DAC), a government-affiliated research and development organisation, has selected HCL Infosystems to validate the AI chip’s design and performance, based on a tender floated on the Government e-Marketplace (GeM) platform.

“Following this, C-DAC will work jointly with HCL to validate the performance of the chip, and the same will then be tested across multiple systems and computing architecture. Our eventual goal is to offer the chip to run on domestic servers and IT infrastructure, and power public services and AI use cases,” the first of the two officials cited above told Mint, both speaking on condition of anonymity.

C-DAC’s expanding role
C-DAC was set up in 1988 under the ministry of electronics and information technology (Meity) to develop indigenous supercomputers after India was denied access to such technology by the US.

“Work on the project has been ongoing for some time now, and the research grants allocated as one of the pillars of Semicon 2.0 will be used in similar projects where India can take ownership of a chip patent. Only then will we truly unlock the kind of value addition that will give us greater independence and worth in the global value chain,” Vaishnaw said.

To be sure, Vaishnaw said private startups would be encouraged to emulate C-DAC’s AI chip development. Funding for the government-backed research organisation, however, will not come from Semicon 2.0, but through the National Supercomputing Mission.

“C-DAC will play a very crucial role in the development of India’s own semiconductor intellectual properties (IPs),” the minister said, adding that incentivization of research and development will be a key part of the new semiconductor mission going forward.

Speaking with Mint, S.D. Sudarsan, executive director of C-DAC Bengaluru, acknowledged progress around work on the AI inference chip, which is used to execute trained AI models on new, real-world data.

“Meity has been highly supportive of all the work that C-DAC has pursued around indigenous silicon architecture, and work around the indigenous AI chip is progressing as per the expected timeline. Early trials have proven to be successful, and we’re now in process of taking the project to the next level,” Sudarsan said.

Beyond the first chip
The inference chip in question is akin to US giant Nvidia’s graphic processing units (GPUs), or other chips such as Google’s tensor processing unit (TPU) and Nvidia-acquired Groq’s language processing unit (LPU). With the indigenous chip, India is looking to have a homegrown patent design that would make it less vulnerable to US export control regulations, which have in the past restricted access to Nvidia’s chips.

In June last year, Mint reported that India had begun work on an indigenous, 2-nanometre (nm) GPU for AI workloads.

A key part of Semicon 2.0 will be served by C-DAC, the officials cited above confirmed. Alongside creating its own AI chip for Indian workloads, C-DAC will “also be the nodal agency for implementing the design-linked incentive scheme,” said Amitesh Sinha, additional secretary at Meity and chief executive of India Semiconductor Mission.

Now, with the goal of developing an indigenous AI chip in sight, C-DAC, Vaishnaw and Sinha said, will play a key validation role in identifying potential Indian chip designs that can support global workloads. LiveMint