India’s foray into neuromorphic computing and the AI ​​revolution

Artificial intelligence (AI) will create one of the largest market opportunities in history, with potential value estimated at between $3.5 and $5.8 trillion. Capturing a significant share of this market could redefine national economies and act as a powerful growth engine for decades to come. For India, leveraging AI is key to achieving the vision of a Viksit Bharat by 2047.

While AI has long been a subject of fascination, it has also seen cycles of breakthroughs and disappointments. A closer look reveals a critical shortcoming: these breakthroughs come with enormous energy demands and costly, time-consuming training procedures. If nothing changes, forecasts suggest that AI energy demands could exceed global energy production by 2035, with profound economic and environmental consequences. This will require a leap in computer hardware that could be significantly more energy efficient than what we have now.

Why is this leap necessary? It comes down to the aging von Neumann architecture, the blueprint for all computers of the past 60 years. In this model, computing power and memory are separated, which slows down operation and consumes energy. For tasks that require billions of calculations per second, such as those used in AI, von Neumann’s design has become a major bottleneck. What’s worse is that the data we generate and use in AI systems is often stored by large companies, raising privacy concerns.

The solution may be closer than we think – in our own heads. The human brain, which weighs less than two kg and uses only 20 watts of energy, is capable of performing billions of actions per second while seamlessly storing and processing information in the same place. This extraordinary efficiency has inspired a new approach to computing, one modeled after the brain’s neural networks.

The concept of brain-inspired computing is not new. In the 1980s, visionary American engineer Carver Mead laid the foundation for what could become the future of computers. Fast forward to the 2010s, industry giants like Intel and IBM have reignited interest in brain-like computing. With advanced manufacturing technologies at their disposal, these companies sought to mimic the brain’s learning processes using traditional binary transistors and software-driven systems. It should come as no surprise that this brute-force approach backfired.

The lesson was clear: to get anywhere close to the brain’s computational efficiency, we must reengineer computers with new circuit elements that can learn and adapt, such as biological neurons and synapses. We also had to rethink the entire computer architecture and move beyond the limitations of Von Neumann systems, where memory and processing are separated.

The race to develop brain-inspired computers isn’t just about mimicking the brain’s processing power; it’s about doing this with the same energy efficiency and compactness that makes the brain so remarkable. The question is: can we build machines that are as smart and efficient as the human brain? The challenge lies in creating computer systems that can store information in thousands of states and, like the brain, operate on the edge of chaos.

In a groundbreaking study published in Nature, a team led by Dr. Sreetosh Goswami from the Indian Institute of Science, Bengaluru, has invented a revolutionary molecular neuromorphic platform capable of storing and processing data across as many as 16,500 states – leaving traditional transistor-based computers. operating in just two states, far behind. By harnessing the dancing of ions in a molecular film, the team created a system that mimics the brain’s complicated data processing method. The molecules and ions that wiggle in the film generate a host of unique memory states. Each movement was assigned to a separate electrical signal – essentially a computer recording thousands of computing states that excel in both energy efficiency and space-saving potential.

The breakthrough doesn’t stop there. In a stunning technological leap, the team used their molecular platform to recreate NASA’s iconic Pillars of Creation image from the James Webb telescope on a simple tabletop setup. Furthermore, they achieved this feat 4,000 times faster and with 460 times less energy than a traditional computer would require.

With a precision of 14 bits, equivalent to 16,384 analog levels, this chip can transform fields ranging from artificial intelligence (AI) to scientific computing. Imagine training complex AI models, such as Large Language Models (LLMs), directly on personal devices such as laptops and smartphones – a process that currently relies on massive server farms and invasive collection of personal data by large companies. This invention could bring AI processing to individual users, providing unprecedented data privacy and democratizing access to advanced AI tools. This is perhaps one of the most disruptive computing innovations to emerge from India, with the potential to position the country at the forefront of global technological progress.

This article was written by Brainerd Prince, Director of the Center for Thinking, Language and Communication, Plaksha University.