They were also able to do this at a fraction of the time and energy that traditional computers would need.
Loads of computing power is needed to train Language Models (LLMs) which are an integral part of Machine Learning (ML) and Artificial Intelligence (AI) platforms such as ChatGPT, Midjourney and DALL-E are all trained on LLMs. Researchers at the Indian Institute of Science (IISc) have developed a brain-inspired analogue computing platform capable of storing and processing data in an astonishing 16,500 conductance states within a molecular film. This breakthrough can bring complex AI tasks to personal devices.
Generally in digital computers data storage and processing are limited to just two states. “Neuromorphic computing has had its fair share of unsolved challenges for over a decade. With this discovery, we have almost nailed the perfect system a rare feat,” said Sreetosh Goswami, Assistant Professor at the Centre for Nano Science and Engineering (CeNSE), IISc, who led the research team.
The newly developed platform can perform complex tasks on personal devices like laptops and smartphones. “These developments are currently restricted to resource-heavy data centres, due to a lack of energy-efficient hardware. With silicon electronics nearing saturation, designing brain-inspired accelerators that can work alongside silicon chips to deliver faster, more efficient AI is also becoming crucial,” read the IISc release. The platform developed drastically cuts down both, time and energy involved, making calculations a lot faster and easier.
“This project brought together the precision of electrical engineering with the creativity of chemistry, letting us control molecular kinetics very precisely inside an electronic circuit powered by nanosecond voltage pulses,” explained Sreebrata Goswami, Visiting Professor at CeNSE.
Tapping into these tiny molecular changes, allowed the team to create a highly precise and efficient neuromorphic accelerator, which can store and process data within the same location, similar to the human brain. Such accelerators can seamlessly be integrated with silicon circuits to boost their performance and energy efficiency.
The team also turned this scientific discovery into a technological feat. They were able to recreate NASA’s iconic ‘Pillars of Creation’ image from the James Webb Space Telescope data – created by a supercomputer using just a tabletop computer. They were also able to do this at a fraction of the time and energy that traditional computers would need.
The researchers believe that this breakthrough could be one of India’s biggest leaps in AI hardware, putting the country on the map of global innovation. Navakanta Bhat, Professor at CeNSE said, “In the context of the India Semiconductor Mission, this development could revolutionise industrial, consumer and strategic applications. The national importance of such research cannot be overstated.”
source/content: newindianexpress.com (headline edited)