Towards Sustainable Electronics: Building Brain-Inspired Computer Chips for Next-Generation AI 

by  
Rachana Bhattacharjee

India in Focus 

Towards Sustainable Electronics: Building Brain-Inspired Computer Chips for Next-Generation AI 

—Rachana Bhattacharjee, based on an exclusive interview with Prof Manan Suri 

 

Artificial intelligence (AI), data mining, deep learning modern computer technologies. 
Futuristic Cyber Technology Innovation. 
Brain representing artificial intelligence with printed circuit board ( 
Image for representational purposes only. 
Image credits: Shutterstock 

Today we, as a civilisation, are generating an enormous amount of data—trillions of bytes every day. Every time somebody carries their smartphone from one room to another or browses the internet, they generate some data. And this is in addition to all the information we’re willingly putting out on the internet.  

Over the last two decades, there has been an explosion of information and the electronics that hold this information. However, our current technology falls short when it comes to making sense of this data. What we need is more advanced forms of software and hardware, like artificial intelligence (AI) and nanoelectronics, that together can receive data, synthesise it, and make decisions similar to how one of nature’s best data processors, the human brain, would.  

Fortunately, research to achieve such technology is rapidly progressing globally today. Prof Manan Suri’s work at IIT Delhi is part of this effort to build this next-generation technology.  

Mimicking the Human Brain at IIT Delhi 

The current electronic systems in semiconductors have two parts: one for storing data—the memory—and the other for computing it—the processor. They have traditionally remained isolated in computer hardware. For instance, our home computers and smartphones have a hard disk to store information and a Central Processing Unit (CPU) to process it. 

To mimic the human brain, one requirement is to fuse this hardware. The information must be processed at or close to its storage location.  

At IIT Delhi, Prof Manan Suri’s team has discovered a novel means of achieving this by turning a property of emerging Non-Volatile Memory (eNVM) that is generally considered a computational flaw into a solution.  

Non-volatile memory systems are those that can store data even after the power source is cut off, like a pen drive, as opposed to a volatile memory system, like Random Access Memory (RAM), where the information is lost when the computer is shut down. Over the past few decades, as electronic hardware systems advanced, NVM chips became increasingly compact. They have now reached the nanoscale in the form of eNVMs.  

Now, the physics of a nanoscale chip makes its behaviour probabilistic and difficult to control, which is an engineering challenge when it comes to traditional electronics that compute data in binaries of 0 and 1. Yet, this very irregularity allows for the processing of information in multiple states between the 0s and 1s and makes the chip ideal for building devices capable of mimicking the complex learning and memory functions of the human brain.  

Neuromorphic Computing for Edge AI 

Using the inherent variability of eNVMs, Prof Suri’s team has devised large-scale nanocircuits that mimic some of the neuro-biological architectures of the human brain. In effect, the memory becomes intelligent in these chips, contributing in real-time to data processing and analysis. The elimination of a back-and-forth of data between the processor and memory units during a computation, in addition to reducing the resources needed to implement storage and computation, results in much faster data processing with much greater energy efficiency.  

 

A figure demonstrating the accuracy of predictions by an eNVM-based AI chip developed and deployed in collaboration with the Israel-based startup Weebit Nano by Prof Manan Suri’s team. | Image credits: IIT Delhi 

This work by Prof Suri and his team lies at the intersection of computational neuroscience, computer science, and electronics and is a global first in many ways. It opens doors to creating processors capable of taking AI to the next level. Where can their principle be applied in real life today? To find out, they incubated a startup called Cyran AI Solutions.  

Cyran AI Solutions: A Deep Tech Startup at IIT Delhi 

Since 2018, their constantly evolving fundamental research at the university has laid the groundwork for the wide variety of user-specific AI solutions devised at Cyran. These are combined hardware–software solutions that process data at or close to where it is collected without connecting to the cloud.  

For instance, in 2019, the team developed a hyperspectral image classification algorithm for low-power neuromorphic hardware on remote-sensing drones. After their preliminary studies showed promising results—an accuracy of close to 97% in a recognition time of 18.4 seconds, with an energy consumption of only about 10 μJ—it was deployed overseas in the Republic of Korea and was widely appreciated. This was a first-of-its-kind technology in terms of computational speed, energy efficiency, and the fact that it could be placed on the drone itself, eliminating the need to process the data separately and taking the functionality of such drones beyond data collection while reducing overall cost.  

In 2020, the team at Cyran developed another first-of-its-kind edge AI solution, this time in a completely different sphere of interest: education. Known as BUDDHI (Build, Understand, Design, Deploy, Human-like Intelligence), this DIY kit enables a child or people from outside the AI field to learn its basics and build simple AI systems. The kit includes an AI computing engine, AI training/inference applications, real-world AI actuation circuit boards, and a high-quality AI handbook, forming a learning ecosystem. The goal is to make the subject area of AI more accessible and intuitive for people, removing the commonly held perceptions of complexity that come with it. These indigenous kits have been a success in many educational institutions nationally. Hopefully, along with the changing times, as AI increasingly penetrates our lives, so will our understanding of the technology.  

These and other solutions, such as those for sensors, security, biometrics, facial recognition, and geospatial image analysis, have earned the team at Cyran national innovation awards such as the National Award for Technology Startups by the Technology Development Board, Department of Science and Technology, the iDEX DISC Award from the Ministry of Defence, and Raksha Mantri’s Excellence Award for Defence and Aerospace. They also won the recent Dare to Dream contest organised by the Defence Research and Development Organisation for neuromorphic vision sensors.  

Cyran’s products might appear vastly different, but the underlying concept is the same—delivering highly customised and precise edge AI solutions for users.  

 

Prof Suri and his team receiving the Raksha Mantri’s Excellence Award for Defence and Aerospace innovations. 
Image credits: Cyran AI Solutions 

Innovations Through Collaborations 

Many of these innovations were collaborative efforts between Prof Suri’s team at IIT Delhi and other national and international companies and institutes. Collaborations, when they involve complementary knowledge exchange and foster new capacity building on both sides, are crucial for innovation.   

For instance, a lot of the team’s fundamental research at IIT Delhi is conducted in collaboration with institutions, foundries, and startups in Europe, the US, and Taiwan. The eNVM-based semiconductors for security applications were the product of a collaborative project between IIT Delhi, IIT Bombay, and Semicounductor Laboratory, Mohali, supported by the Office of the Principal Scientific Adviser to the Government of India. In collaboration with Israel-based semiconductor startup Weebit Nano, the team is helping apply Weebit’s unique eNVM technology to computer chips for AI.  

Intelligent Memory is the Basis of our Future 

Neuromorphic chips are the basis of semiconductors that become platforms on which edge AI can run. These enable processing speed, energy efficiency, and cost savings on a level that finally makes the possibility of ubiquitous edge AI real. If these chips can be produced indigenously, we can harness the potential of the enormous amounts of data we have today to a much greater extent.    

Today, while a human brain computes extremely complex data across billions of neuronal junctions within seconds, using only about 20 Watts of power, our finest supercomputers require many gigawatts of power, elaborate infrastructure, and considerable time to process much less information. Imagine a future where this is no longer the case. Imagine a future where the devices in our hands make sense of data as well as our brains can while consuming as little power as our brains do. Now, that will be a truly sustainable future for electronics.  

With the work at IIT Delhi and Cyran, Prof Suri’s team has taken the first steps towards this future.  

About Prof Manan Suri  

Prof Manan Suri is an Associate Professor in the Department of Electrical Engineering at IIT Delhi, Principal Investigator in the NVM and Neuromorphic Hardware Research Group at IIT Delhi, and Founder of Cyran AI Solutions. 

About the author 

Rachana Bhattacharjee is an author, creative lead, and one of the countless chroniclers of the information age. 

topbutton

Back to Top