(Published Claws 30 Apr 2016 )
..And I had an opportunity to grow from the time where we couldn’t make a single silicon transistor to the time where we put 1.7 billion of them on one chip!
Gordon Moore, Cofounder Intel
Last year Kris Gopalakrishnan pledged $ 50 mn at IISc and IIT Madras on research that seeks to model next level computing based on the functioning of the Brain. Neuromorphic engineering is an emerging interdisciplinary field that involves designing sophisticated devices based on the complex neural circuits of the brain. It uses principles of the nervous system for engineering applications to achieve a better understanding of computations occurring in actual biological circuits and utilize the unique properties of biological circuits to design and implement efficient engineering products. Neuromorphic chips aim to mimic the massive parallel computing power of the brain, circumvent the size limitations of traditional chips, and consume less power. It is also predicted that such chips could adapt in response to stimuli. As a technology demonstrator, P. Merolla et al  at IBM have developed a 5.4-billion-transistor chip (TrueNorth) with 4096 neurosynaptic cores interconnected via an intra-chip network that integrates 1 million programmable spiking neurons and 256 million configurable synapses. With 5.4 billion transistors, occupying 4.3-sq cm area TrueNorth has ∼428 million bits of on-chip memory. In terms of power, consumption where a typical central processing unit (CPU) consumes 50 to 100 W per sq cm the TrueNorth’s power density is 20 mW per sq cm only. This qualifies it to be a good candidate for ushering in green technology in to computing. However, for purposes of clarity TrueNorth is not a brain, it is inspired by the brain and mimics some functions of the brain to carry out computations.
Market for Neuromorphic Chips
The main factors, which have driven research and development of neuromorphic chips, are tremendous demand for data and data analytics, miniaturization of sensors, ingress of Artificial Intelligence into software of almost all intelligent machines and high cost of further miniaturization of integrated circuits. These factors have spurred the demand and growth of the market for neuromorphic chips, which is expected to grow at a CAGR of 26.31% between 2016 -2022. One of the key areas where such systems would need break-through research would be in design of algorithms since biological systems autonomously process information through deep learning whereas any human designed chip or system would be limited by human designed algorithms. The applications areas currently comprise sensors in military as well as medical fields.
Militaries today are coping up with an exponential increase in the amount of data from a wide variety of sensors. Unprecedented data collection has severely strained the limited available bandwidth for military use. The data needs to be processed, as close to the sensor as possible before further transmission therefore sequential computational techniques with their large size and power requirements are not very efficient in this regard. NeuroSynaptic chips can carry out this parallel task much more efficiently.
DARPA had initiated a project called Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE), in 2008 and had contracted it to IBM and HRL. It has funding of over $ 100 mn. The aim of SyNAPSE is stated ‘to build an electronic microprocessor system that matches a mammalian brain in function, size, and power consumption. Further, it should recreate 10 billion neurons, 100 trillion synapses, consume one kilowatt, and occupy less than two liters of space’.
The US Army has projected a requirement for a high-performance, low-power bio-inspired parallel processor. This would be integrated in to cognitive communication systems and image processing platforms on unmanned vehicles. The project is being undertaken by Physical Optics Corporation (POC) under their BRAINWARE processor program.
The U.S. Air Force has projected a requirement to develop a new class of advanced, wide field of view (WFOV) imaging sensors that sample the radiation field in multiple modes: spectral, temporal, polarization, and detailed object shape. These multimodal sensors are for deployment on high altitude ISR functions of drones. Scaled down versions are required for use with autonomous micro-air vehicles (MAV) for guidance, navigation, and control. Two types of bio-inspired multimodal sensors, one operating in the visible wavelength regime, and the other operating in the infrared wavelength regime are being developed by The Spectral Imaging Laboratory (SPILAB) in collaboration with the University of Arizona. Both sensors will have a neuromorphic processing capability based upon visual brain areas of insects and crotalid snakes.
It is apparent that neuromorphic chip based computational systems scalable to the capabilities of the human brain are a clear possibility provided an all-round research and development effort is synergized in hardware, software, architecture, and simulation & understanding of functioning of the brain. The neuromorphic chips as well as quantum computing have ushered in a paradigm shift from the focus on microchips to that of the system as a whole.
In the ultimate goal of mimicking the human brain, it is likely that development of artificial brains of smaller species or specific parts of the human brain may turn out to be more enchanting purely from a commercial point of view. The impetus to the rapid development in neuromorphic systems would be provided by the availability and applications of such systems for large-scale commercial utilization.
 Computational power efficiency for biological systems is 8–9 orders of magnitude higher than the power efficiency wall for digital computation;