Presidency research scholar chosen for ‘Top Master in Nanoscience’
• Your research interest is in the field of Neuromorphic Computing. How is it related to the ever-expanding arena of nanoscience?
Neuromorphic Computing or ‘neuron-like computing’ is inspired from how our nervous system and brain do different kinds of computations which are essential to life. There are certain tasks in which the human brain is much better and more energy efficient than even the best supercomputers today. The idea is to achieve those too on hardware, through Neuromorphic Computing. This is a paradigm shift in the conventional computational landscape in the sense that it is biologically inspired. This makes it a unique discipline, wherein we are inspired by the techniques of mother nature and try to implement them in the context of our own benefit. Another exclusive feature of it is its multidisciplinary nature. People from varying disciplines like Physics, Chemistry, Biology, Computer Science, Artificial Intelligence, Engineering are putting together their heads for this.
• How can very-Large-Scale integration (VLSI) systems be utilized to simplify, rather evolve the complex process of human brain-mapping?
Human brain-mapping is a task very much distinct from Neuromorphic Computing. In brain-mapping, neuroscientists try to simulate every part of the brain exactly and to understand how it works whereas in Neuromorphic we are only interested in utilizing the basic principles through which a neuron functions. The brain has 86 billion neurons which is a very big number. Thus, to achieve comparable computing ability, a highly dense chip architecture is required which can be achieved by VLSI. Components analogous to neurons and synapses (connection points between neurons) need to be fabricated in huge numbers to realize tasks as complex as the ones we are solving right now.
• Is sparse coding the only processing the human brain does for pattern recognition, or there are other such neurological procedures too?
How the brain exactly achieves pattern recognition is to a large extent an open question. One of the possible ways by which it might do a pattern recognition task is by sparse coding where the brain looks for familiar features in a sensory input. For example, when we look at a person, our brain analyzes them in terms of some basic components like shape of eyes, presence of a beard/ moustache; and then come to a final conclusion.
• How can memristors be integrated in advanced scientific solutions to enhance the prospects of neurological research?
Nervous systems comprise of a huge number of neurons which are interconnected in a much larger network through connection points called synapses. The strength of connections between neurons is an important component behind learning. If a connection is weakened, then it is somewhat analogous to the suppression of a functionality and if it is strengthened, then it is the other way around. For instance, if we are shifting to a new country we start learning the street names there when we may forget some names from the previous country. This can be thought of as some synapses (responsible for remembering the names of the streets in the previous country) getting weakened and different ones (responsible for remembering the names of the streets in the new country) being activated. Now, these synapses are in principle, just channels through which electrochemical signals pass carrying an information (like a street name) from one neuron to another. If this connectivity is to be weakened, the resistance to the flow of the signal needs to increase. This controlled change in resistance is not easy to achieve in conventional materials and this is the reason why memristors are interesting! Memristors or Memory-resistors are components whose resistance varies depending upon its history. So, operating the memristor in different ways can give rise to different values of the resistance, thus it can learn or forget street names. This is how materials (memristors) can emulate real synapses that enable learning in Neuromorphic Hardware.
• Does the response time of a healthy brain match with that of a computer circuit for a particular activity? Any example?
The response time of computers and human brain heavily depend upon the task at hand. If the task is to evaluate 1234x9876 then the computer is obviously much better at doing it. But if it is a more complicated task like identifying a person in a crowd or showing creative aptitude, then the human brain is way better than the best of computers available today. This bottleneck is due to the intrinsic differences in the internal architecture and functioning of the two systems. The human brain is also superior to a computer in terms of usage of energy. If they both can do the same task taking the same amount of time, then the brain needs energy which is orders of magnitude less than what is needed by a computer. This energy efficiency is another aspect of why we want to achieve Neuromorphic Computing.
• What are the minimum infrastructural requirements for performing research in Neuromorphic Computing? Does India have that provision?
Research in Neuromorphic Computing requires infrastructural support in the fabrication of nanoscale devices, its characterization, and measurement. To fabricate the device appropriate cleanrooms and advanced machines are required. The characterization and measurements also need sophisticated instruments and expertise. Moreover, a healthy overlap of research ideas between people from different disciplines is needed. Even if the first aspect is in there is some premier institutes in India, the multi-discipline approach is lacking. Although in the recent years this bridge is being gapped and today there is some effort in India in this direction too.
• Which industrial sectors are soon going to see a boost in their services due to application of Neuromorphic Computing?
The industries that rely on huge computational power to do cognitive tasks or tasks requiring Machine Learning are sure to see a boost. Computers require much more power to execute pattern recognition or cognition tasks which can be made more efficient with Neuromorphic Computing. Neuromorphic chips like SpiNNaker, TrueNorth, BrainScaleS, Loihi are already showing some promise in this regard. Finally, the Silicon transistor-based industries will also see a change resulting from this as the conventional architecture systems themselves are going to be challenged.
• What are the multidisciplinary aspects of this field and how are they cohesively blended in terms of developing scientific solutions?
This field takes inspiration from Biology, tries to do tasks related to Artificial Intelligence and to do so, uses the Physics and Chemistry of materials. So, the very essence of multidiscipline is ingrained in the subject matter. Scientists from these fields must work cohesively and together, to achieve a goal. The scientific solution to any problem generally relies on a multitude of factors. After the concepts are idealized and the devices are fabricated, it remains a challenge to produce them in huge quantities. Production engineering and circuit designing have a crucial role to play in this aspect. So, in this gigantic puzzle, each piece however different it might be from the other ones- must fit together. Only then is the whole picture complete and the world a happier place.