Although neuromorphic approaches to hardware and software are still nascent, India should explore them to become a key player in the field of emerging technologies.
Image Source: getty
Neuromorphic approaches create hardware and software that is modelled upon biological neural computation and its associated ability to process, understand and react to data in milliseconds while consuming relatively minuscule amounts of power. The theoretical potential of these approaches to maximise efficiency across Artificial Intelligence (AI) applications has garnered interest from several stakeholders, including companies such as Intel and IBM and governments worldwide.
The theoretical potential of these approaches to maximise efficiency across Artificial Intelligence (AI) applications has garnered interest from several stakeholders, including companies such as Intel and IBM and governments worldwide.
In late 2024, researchers at the Indian Institute of Science (IISc) in Bengaluru reported that they had built a brain-inspired computing platform. This aligns with a larger trend that focuses on neuromorphic approaches towards hardware and software. Although the neuromorphic field is relatively nascent, India should pay attention to new developments because increased neuromorphic research could contribute to positioning the country as a global leader in research on emerging technologies.
Two avenues within the field have been at the forefront of exploration. The first is the use of Spiking Neural Networks (SNNs). SNNs are based on neural synapses in the brain and utilise spikes to encode and transmit information. This differs from the Artificial Neural Networks (ANNs) that have contributed to current AI applications. That said, SNNs may not be well-suited to conventional hardware. Consequently, another area of research is the development of neuromorphic chips, which could have the capability to support SNNs.
Traditional chips store and process data in separate places. This has created what is known as the von Neumann bottleneck, wherein the separation of memory and computing creates the need for tasks to be completed sequentially. In contrast, neuromorphic chips aim to maximise energy and computing efficiency by removing the gap between memory and processing. Energy efficiency becomes particularly important in the context of projected energy demand. For instance, the International Energy Agency expects that powering AI, cryptocurrency, and data centres will require one thousand terawatt hours by 2026. Neuromorphic approaches to hardware and software could also benefit edge applications that require in-memory processing and real-time insights, such as autonomous vehicles, robotics (particularly prosthetic limbs), and banking and financial services.
Nevertheless, neuromorphic chips and SNNs are not expected to replace other technologies (such as ANNs, quantum technologies or conventional chips) because efficiency gains may only be applicable in certain use cases. Research suggests, for instance, that conventional chips are best suited for performing precise calculations, while quantum technologies have applications within encryption, communication, and quantum sensing that neuromorphic chips do not compete with.
Neuromorphic chips and SNNs are not expected to replace other technologies (such as ANNs, quantum technologies or conventional chips) because efficiency gains may only be applicable in certain use cases.
Additionally, the exploration of neuromorphic approaches is constrained by certain issues. Firstly, building neuromorphic hardware and associated algorithms requires rare, cross-functional expertise across several fields, including neuroscience, circuit design, physics and computer science. Secondly, the advent of neuromorphic computing is complicated by our evolving understanding of the human brain's functioning. Finally, there are also significant issues with replicating the level of complexity of the human brain, which has over 100 trillion synaptic connections and 100 billion neurons. Accordingly, neural networks show increased performance with more neurons and synapses, but it is difficult to obtain the numbers required for brain-like functioning. Further, even if the neurons and synapses available increase as a result of combining chips, energy efficiency may be lost on account of interconnections. Some researchers also question whether it is due to a lack of clear benchmarks and the relative nascency of neuromorphic hardware.
This introduces another problem. Conventional hardware is not well-equipped to handle neuromorphic research, which requires the creation of neuromorphic-compatible hardware. This will inevitably require capital as well as time. Additionally, it has been proven difficult to train SNNs because their characteristic spikes are mathematically discontinuous, though several research groups are making progress on this front. There may also be losses of precision with SNNs, which is why researchers are experimenting with combining ANNs and SNNs. Consequently, there is disagreement between neuromorphic researchers with respect to the feasibility and practicality of creating architectures that replicate the human brain precisely, as opposed to creating architectures that are simply inspired by the brain.
Despite these challenges, stakeholders around the world are increasingly invested in exploring the feasibility of neuromorphic approaches. At the intergovernmental level, for instance, in an attempt to better grasp the functioning of the human brain, the European Union launched the Human Brain Project in 2013, which has since released two neuromorphic systems, BrainScaleS and SpiNNaker. In 2024, the European Union (EU) and the Republic of Korea also announced a partnership that focused on the advancement of neuromorphic technologies, with particular reference to semiconductors.
Hala Point can work with conventional AI workloads and is powered by Intel's Loihi 2 chip, which can support other neuro-inspired algorithms.
Several private actors are also researching neuromorphic approaches and preparing for their integration into the market at various levels. Intel recently released a statement announcing that its Hala Point neuromorphic system (which, with its 1.15 billion neurons, is reportedly the largest in the world) will be used to support future brain-inspired AI and research at Sandia National Laboratories. Hala Point can work with conventional AI workloads and is powered by Intel's Loihi 2 chip, which can support other neuro-inspired algorithms. In addition to its chips, the company has also initiated the Intel Neuromorphic Research Community, which aims to work with government facilities, academia and other companies to further neuromorphic research. Additionally, IBM is researching materials that can be used to create neuromorphic architecture, such as ferroelectrics. The company has also launched its brain-inspired NorthPole chip, which has placed memory and computing on the same device. Consequently, the chip is reportedly capable of performing AI inference faster than all other available chips. Similarly, Samsung’s neuromorphic research is all set to move forward, and it plans to expand its neural processing architecture into neuromorphics over the next decade. Other companies are assessing the feasibility of neuromorphic approaches for specific use cases. The Mercedes-Benz Group, for instance, is a part of multiple projects, including NAOMI4Radar and a research collaboration with the University of Waterloo, which aim to research neuromorphic computing to improve automated driving.
Developments at IISc earlier this year have demonstrated that there is immense potential for neuromorphic research in India, and global interest in the field has shown that it is important for India to encourage further indigenous research. For one, India has the opportunity to present itself as a potential research partner to other states working on neuromorphic technology, including the US and EU. This could pave the way for further bilateral and multilateral partnership opportunities, thereby placing India at the forefront of global research and development in emerging technologies. Additionally, research on neuromorphic chips that yield tangible results would likely contribute to India having an advantage in intellectual property rights. This would help India position itself as a key global player in technology innovation and manufacturing. Further, increased research could allow the country to prepare for a future boom in neuromorphic technologies. For instance, Indian companies researching neuromorphic chips before they become mainstream could identify specific use cases for neuromorphic approaches. They would also be likely to create a more nuanced understanding of how neuromorphic approaches could impact technology convergence within various industries.
Pranjali Goradia is a Research Intern at the Observer Research Foundation
The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.
Pranjali Goradia is a Research Intern at the Observer Research Foundation ...
Read More +
PREV
