Time is ticking for this brain-inspired chip poised to change the game for AI

Bengaluru, Apr 10 (PTI) Even as deep tech debate is raging in India, triggered anew by Union Minister Piyush Goyal’s call to “focus on real businesses”, a leading scientist behind a groundbreaking innovation — a brain-inspired chip that will propel the Artificial Intelligence (AI) to greater heights — is sounding the alarm: time is running out for his project.

“AI is evolving at breakneck speed. If we lose a year, we risk losing agency,” said Sreetosh Goswami, Assistant Professor at the Centre for Nano Science and Engineering, Indian Institute of Science (IISc), Bengaluru.

Goswami’s ‘memristor’ is possibly one of the most written about technologies even during its research phase. His paper published in ‘Nature’ journal sparked a lot of interest in his project, he said.

Short for Memory Resistor, 'memristor' lies at the core of a new class of AI hardware known as molecular neuromorphic computing (MoNC-AI).

For lay people, that would translate as a brain-inspired circuitry that could help AI systems become far more energy-efficient, secure, and intelligent, solving some of the biggest bottlenecks in today’s AI race.

For all its dazzling progress, AI still runs on hardware that’s woefully out of sync with how natural or biological intelligence actually works, said Goswami.

“Silicon chips are phenomenal at precision — they crunch numbers with ruthless accuracy. But perception? Learning? Adaptation? That’s not what transistors were born to do,” Goswami told PTI.

Whereas, biology doesn’t do precise calculations — it feels, learns, adapts, he said.

“So, we asked ourselves: can we build hardware that does the same?” added Goswami.

More than a decade of research later, ‘memristor’, an example of what happens when chemistry meets cognition, made it possible for his team to design a molecular neuromorphic chip.

“It is capable of storing information in 16,500 distinct levels, it’s the most accurate memory hardware ever engineered. And it doesn’t guzzle power like traditional AI chips. In fact, it’s 220 times more energy-efficient than Nvidia’s top-tier GPUs — all while retaining the power to learn and adapt,” said Goswami.

The chip was exhibited at India’s first Nano Electronics Roadshow, held at IISc Bengaluru on March 27. The roadshow was organised by the Ministry of Electronics and Information Technology (MeitY), in partnership with Indian Institutes of Technology from Chennai, Mumbai, Delhi, Kharagpur, and Guwahati.

One would think that with a breakthrough technology like ‘memristor’, and what it could potentially mean to AI, things will proceed faster.

But the ecosystem still has some catching up to do, said Goswami.

“At the stage we’re operating now, I’ll admit, funding isn’t the bottleneck. The government has been incredibly supportive,” he said.

But to scale this into a globally competitive technology, Goswami said they will need a significant push. “Not just from public agencies, but from private stakeholders as well,” he added.

Scaling up, Goswami emphasised, demands more than just innovation too; it requires infrastructure, specifically a semiconductor fabrication facility, or fab as it’s commonly called.

“India still lacks a state-of-the-art fab,” he pointed out. “Right now, we tape out chips abroad and then layer our technology on top. It’s not impossible, but let’s be honest, it’s not ideal.” The real game-changer, Goswami said, would be a domestic fab capable of integrating indigenous technologies directly into existing chip architectures. The upcoming Tata semiconductor foundry has sparked optimism, he added.

“This is exactly how it’s being done in China. Their researchers work hand-in-hand with foundries, embedding their innovations into CMOS processes, sometimes right within university-based fab lines,” he pointed out.

CMOS or Complementary Metal-Oxide-Semiconductor is, in Goswami’s words, “the beating heart” of modern electronics. “We can’t replace it. But we can elevate it. And to do that, researchers need access to the fab floor — to experiment, to iterate and to push boundaries,” he said.

The immediate goal for the team is a “proof-of-concept” chip: a tangible demonstration that this molecular neuromorphic technology is ready for prime time.

Goswami expects it to be ready within two years. “But we’re in a race against time. AI evolves at breakneck speed. If we don’t move fast, we risk losing relevance,” he said.

If everything falls into place, Goswami believes this platform could do more than just transform computing. It could redefine economic priorities and bolster national security.

“Look at where we’re headed: massive data centres that guzzle power, drain water resources, and emit unsustainable levels of CO2. Do we build a nuclear plant for every AI server farm? If this continues unchecked, the next global crisis will be over natural resources,” he said.

What we need now is not incremental improvement, but a homegrown breakthrough, believes Goswami.

“And molecular neuromorphic chip could very well be it,” he added. PTI JR ADB

Source: PTI