IBM builds cognitive chips at Fishkill fab

August 19, 2011 — IBM (NYSE:IBM) researchers have build experimental chips that emulate the perception, action and cognition of brains. The chips aim to drastically lower power consumption and form factors in future "cognitive computers."

The first 2 prototype chips — which contain no biological elements — have been fabbed at IBM’s Fishkill, NY fab and are undergoing tests at IBM’s Yorktown Heights, NY and San Jose, CA research labs. Both prototype cores were fabricated in 45nm silicon on insulator-complementary metal oxide semiconductor (SOI-CMOS) and contain 256 neurons. The digital silicon circuits inspired by neurobiology create a "neurosynaptic core" with integrated memory (replicated synapses), computation (replicated neurons) and communication (replicated axons). One core contains 262,144 programmable synapses; one contains 65,536 learning synapses. They perform simple applications like navigation, machine vision, associative memory, pattern recognition, classification, etc.

IBM’s "neurosynaptic computing chips" recreate the phenomena between spiking neurons and synapses in biological systems through advanced algorithms and silicon circuitry. Cognitive computers will be programmed by experiences, finding correlations, creating hypotheses, and remembering the outcomes. The cognitive system will analyze complex information from multiple sensory modalities simultaneously, dynamically rewiring itself as it interacts with its environment. IBM’s cognitive computing architecture of on-chip lightweight cores create a single integrated system of hardware and software.

Under the Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) multi-year project, IBM combined nanoscience, neuroscience and supercomputing expertise in a cognitive computing initiative with university collaborators. The US Department of Defense’s Defense Advanced Research Projects Agency (DARPA) awarded the team about $21 million in new funding for Phase 2 of SyNAPSE. The IBM team successfully completed Phases 0 and 1.

These brain-inspired chips move beyond "more than half a century" of von Neumann paradigm computer architectures, said Dharmendra Modha, project leader for IBM Research. The cognitive system has no set programming, integrates memory with processor, and mimics brains’ event-driven, distributed and parallel processing.

IBM plans to build a chip system with 10 billion neurons and 100 trillion synapses that uses 1 kilowatt of power, with a 2-liter volume form factor.

Potential applications include vast sensor networks feeding a world water monitoring system for agriculture to tsunamis, grocer shelves that detect bad or contaminated produce, traffic lights that know when driving conditions become dangerous, and endless consumer product uses, said Dr. Modha.

University partners include Columbia University; University of California, Merced; Cornell University; and University of Wisconsin, Madison.

IBM has performed cognitive computing research since 1956 and recently created "Watson" to demonstrate problem-solving functions of advanced computers. For more information about IBM Research, please visit

Throughout 2011, IBM is hosting the IBM Research Colloquia, convening thought leaders at its global labs to discuss technologies of the future and their potential impact on business and society. The first of these colloquia took place at IBM Research – Zurich, and featured a dialogue on Nanotechnology and the Future of Computing with IBM Fellows and Nobel Laureates, Drs. Gerd Binnig and Heinrich Rohrer, and talks by Prof. Dr. Achim Bachem of the Julich Research Center on 21st Century Supercomputing; Prof. Dr. Karlheinz Meier, Kirchhoff Institute for Physics at Heidelberg University on Brain Inspired Computing and Prof. Dr. Daniel Loss of the University of Basel on Quantum Computing.

Chips are also in development that incorporate brain neurons on silicon. Read IMEC’s Patterning neurons-on-chip devices using microcontact printing


Easily post a comment below using your Linkedin, Twitter, Google or Facebook account. Comments won't automatically be posted to your social media accounts unless you select to share.