by Introduction
Neuromorphic chips aim to harness the power of the human brain by mimicking its architectural design and functionality at the circuit level. These neuromorphic or “brain-like” chips may have the potential to revolutionize computing by achieving higher performance than traditional computers for certain tasks like pattern recognition and cognitive abilities that come naturally to humans and animals.
What are Neuromorphic Chips?
Neuromorphic Chips are artificially created electronic circuits that replicate the behavior of biological neurons found in our brain and nervous system. Just like the brain which consists of billions of neurons connected through synapses, neuromorphic chips contain thousands or millions of artificial neurons that communicate through electronic analogues of synapses.
The key features of neuromorphic chips include:
– Massively parallel architecture modeled after the dense connectivity between neurons in the brain.
– Event-driven, asynchronous computation to closely mimic how the brain processes information.
– On-chip learning capabilities through dynamically adaptable synaptic connections between neurons.
– Low-power design suited for edge and embedded applications.
One of the major research efforts in neuromorphic engineering involves reverse-engineering the brain’s neuronal structure and activity down to the biophysical and biochemical level. This reverse engineering feeds into designing and developing neuromorphic hardware that functions in biologically realistic ways.
The First Generation of Neuromorphic Chips
Some of the pioneers who established the field of neuromorphic engineering include Carver Mead who introduced the concept of neuromorphic systems in the late 1980s. This led to the development of the first generation analog neuromorphic chips in the 1990s.
One of the earliest neuromorphic chips was the Adaptive Silicon Neuromorphic Processor (ASNP) developed by Anthropic in 1995. It contained over 600 analog neurons and over 60,000 synapses on a single chip. The ASNP demonstrated basic perceptual capabilities like visual processing, but the chip architecture was limited.
In the late 1990s, researchers at IBM including Kwabena Boahen created silicon retinas and silicon cochleas capable of visual and auditory processing. The IBM group later developed the first asynchronous spike-based neuromorphic systems with custom made chips called “Neurogrid” and “TrueNorth” that demonstrated deep learning with spiking neural networks.
These early prototypes paved the way for specialized event-driven “brain-inspired” systems, but real-world applications were still far away due to constraints in chip design, learning rules and software support at the time.
Neuromorphic Chips in the 21st Century
Since the 2000s, there has been a renewed interest in neuromorphic engineering driven by the limitations in Moore’s Law and von Neumann architecture. This led to a new generation of digital neuromorphic chips with vastly improved capabilities.
In 2005, researchers at Stanford university developed an experimental chip called “Neurogrid” containing over one million simulated neurons and over 250 million synaptic connections. This breakthrough demonstrated the promise and complexity possible with modern VLSI implementations.
A major milestone was the Intel Corporation’s “Loihi” neuromorphic research chip launched in 2018. Loihi delivered over 130,000 neuro-synaptic cores on a single chip with event driven online learning on par with the human brain. Its goal was accelerate research on specialized deep learning applications.
In 2021, IBM launched “TrueNorth 2”, its second generation chip based on its original TrueNorth architecture. TrueNorth 2 contained over 4.9 billion synapses and nearly 5 million neurons on a single chip, significantly more than its predecessor. It showed applications in computer vision, data analytics and speech recognition.
The Human Brain Project funded by the European Union continues its work on building various large-scale neuromorphic systems including “SpiNNaker” developed by University of Manchester with over 1 million ARM cores. SpiNNaker has been used to model brain regions and complex cognitive functions.
Future Outlook and Applications
While neuromorphic chips have come a long way, we are still in the early stages of learning to build computing systems as complex and power-efficient as the human brain. Continued progress relies on overcoming multiple technical and conceptual hurdles:
– Developing compact models of biology down to the molecular and genetic level for further optimization.
– Designing sophisticated learning rules that combine supervised and unsupervised learning paradigms.
– Efficient simulation of large-scale neural networks with millions of heterogeneous neural components.
– Improved interface between neuromorphic hardware and traditional von Neumann systems.
– Self-organization and self-repair capabilities exhibited by biological brains.
As neuromorphic engineering advances, potential applications include:
– Visual object recognition for robotics and self-driving cars.
– Expert systems for medical diagnosis and financial data analysis.
– Cognitive assistants with natural language processing and context awareness.
– Edge and embedded AI for applications like predictive maintenance and personalized medicine.
– Brain-computer interfaces and biohybrid systems combining biological and electronic components.
Neuromorphic chips represent a radically different computing paradigm inspired by neuroscience for creating intelligent systems. While neuromorphic technology is still in its infancy, it holds promise to revolutionize AI, computing and potentially inspire new developments in other disciplines like material science and cognitive science. With continued innovation, this new generation of “brain-like” chips may help build powerful computing systems that match or even surpass human cognitive capabilities in the future.
*Note:
1. Source: Coherent Market Insights, Public sources, Desk research
2. We have leveraged AI tools to mine information and compile it
