Originally posted by: Th. Mühge
The emerging field of "brain inspired computing" has the potential to revolutionize the computer industry. So why is it so promising to design new computers inspired by the human brain?
Current computer processors are based on the “Von Neumann”-architecture that originates from 1945. This design is extremely powerful for analytic tasks like calculations, regression analysis etc. State of the art computers can execute tasks with ultra-high speed of up to 5 Ghz or an equivalent of 0.2 nano-seconds per operation. In addition they can access a massive amount of information stored in their memory. However, there are major disadvantages of this current computer technology:
High energy consumption due to continuous CPU activity and technology-driven losses
Limited possibility for future major technological improvements
Now we’re leaving the classic computer technology and focusing on the human brain instead. Let’s start and try to compare current computer power with the human brain: Following Chris Westbury, the human brain consists of 100 billion (10¹¹) brain cells called neurons. Each neuron can send out about 100 signals per second, which equivalents to about 1 signal every 10 milliseconds. This seems to be rather slow compared to present processor speeds, but the neuron is not acting alone, but in a large and complex network. The brain functionality is based on this network and the interconnects (Synapses). Each Neuron is connected to approximately 1000 other neurons. So doing a simple multiplication of previous values, we can estimate the total number of operations per second by 10¹⁶ or 10 Million Billions. This in fact is the number of operations current supercomputers are able to execute. Now in order to perform this tasks the brain consumes approximately 20 W of energy (lightbulb), compared to some MW consumed by a supercomputer system (electric energy consumption of a city district).
This is a quite simple approach, but I think it outlines the enormous power of such networks very well.
In turn two major research activities toward a new IT technology arise:
Rebuilding the brain in "classical" computers by simulations. Due to the complexity of the task supercomputers are required.
Transferring the research findings into new "brain inspired" computer hardware
Let me provide a short introduction into "brain inspired" hardware: This introduces a completely new processor concept. The current approach transfers neurons and the interconnecting network into 2 dimensional silicon hardware. IBM is using 28nm CMOS technology for its "brain inspired" SyNAPSE Chip (Systems of Neuromorphic Adaptive Plastic Scalable Electronics).
This chip is powered by 1 million neurons and 256 million synapses. It is the largest chip IBM has ever built at 5.4 billion transistors, and has an on-chip network of 4,096 neurosynaptic cores. Yet, it only consumes 70mW.
I travelled to the Forschungszentrum Jülich two months ago in order to learn more about "brain inspired computing" from top international researchers. As a result of the visit the Jülich team and IBM prepared the following video jointly together. It provides an introduction to this new IT field and the underlying challenges to simulation software and IT infrastructure.
Processing Speed of the human brain
IBM Neurosynaptic Chip
IBM Brain Like Chip
Science: A million spiking-neuron integrated circuit with a scalable communication network and interface
Decorrelation of Neural-Network Activity by Inhibitory Feedback
The Correlation Structure of Local Neuronal Networks Intrinsically Results from Recurrent Dynamics