DARPA Synapse phase 2 targets integrated neuromorphic chip

The status of goals of the DARPA synapse brain emulation project. It appears on track for human scale brain emulation by 2019. It is funded at $30 million per year. Europe could kick off a $1.6 billion human brain emulation project with a target for 2024. It seems like there will be a human brain emulation race. China and other countries likely to join.

The vision of the Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) program is the development of biological-scale neuromorphic electronic systems for autonomous, unmanned, robotic systems where humans are currently the only viable option. The successful development of this technology will revolutionize warfare by providing intelligent terrestrial, underwater, and airborne systems that remove humans from dangerous environments and remove the limitations associated with today’s remote-controlled robotic systems. Applications for neuromorphic electronics include not only robotic systems, but also natural human-machine interfaces and diverse sensory and information integration applications in the defense and civilian sectors. If successful, the program will also reinvigorate the maturing microelectronics industry by enabling a plethora of computer and consumer electronics applications.

FY 2010 17.025 million
FY 2011 27.608 million
FY 2012 31.000 million

HRL Labs, HP and IBM have gotten funding

In Phase 2, the team will pursue developing a fully integrated neuromorphic chip. “The chip design will ensure that the hardware is scalable to support very large-scale neuromorphic architectures,” Srinivasa said. The team will also demonstrate more complex behaviors in the areas of visual perception, planning and decision making and navigation via integration of brain architecture with a virtual environment.

By the end of 2012
– Design and verify a hardware neural system of ~10 billion synapses and ~1 million neurons.
– Demonstrate a chip fabrication process and development plan supporting ~10 billion synapses per square centimeter and ~1 million neurons per square centimeter

100 chips at the end of 2012 would achieve 100 million neurons and 1 trillion synapses

In 2015 the goal is a prototype chip simulating 10 billion neurons connected via 1 trillion synapses. The device must use 1 kilowatt or less (about what a space heater uses) and take up less than 2 liters in volume. 100 of the systems would have 1 trillion neurons and 100 trillion synapses and would be about the complexity of the human brain.

IBM research suggests that a full-scale model of the human brain—which has 20 billion neurons connected by about 200 trillion synapses—could be reached by 2019, given enough processing power. It would be a hardware model. This does not indicate the actual intelligence that would be in the system. It also does not specify the quality of the neurons and synapses that are part of the system.

Still being at human brain scale would be interesting and it would be interesting to see what could be possible and what will be learned. Refinement to better neurons and synapses could progress in the 2020s.

We covered the summary presentation after the end of the first phase

Henry Markram is in talks with the EU for $1.6 billion in funding through 2024 for his artificial human brain project (through 2024 and using multi-level simulation to reduce computational challenge)

Assuming that both Synapse and Markram are funded through 2024, then the total funding would be over $2 billion for human brain emulation. If there is even the smallest belief in the possibility of success with these projects there will also be comparable scale projects in China, Japan, India, Russia and other countries. It seems that any failure of the Singularity and human level general intelligence will not be from lack of funding, desire or effort. However, it could be like nuclear fusion where large sums are spent on approaches that are flawed.

FY 2010 Accomplishments:
– Developed a brain-inspired neuromorphic architectural design and specification capability.
– Developed software tools to translate neuromorphic designs into electronic implementations using hybrid Complementary Metal-Oxide Semiconductor (CMOS) and high-density electronic synapse components.
– Developed capability to simulate the performance of neuromorphic electronics systems using very large scale computation.
– Developed virtual reality environments intended for training and evaluating electronic neuromorphic systems and their corresponding computer simulations.
– Developed standard testing protocols for assessing the performance of large neuromorphic electronic systems.

FY 2011 Plans:
– Demonstrate all core microcircuit functions in hybrid CMOS electronic synapse hardware.
– Demonstrate a dynamic neural system simulation of approximately one million neurons that shows plasticity, self-organization, and network stability in response to sensory stimulus and system level reinforcement.
– Develop tools to design electronic neuromorphic systems of 100 billion neurons with mammalian connectivity.
– Demonstrate virtual environments with a selectable range of complexity across the cognitive capabilities of small to medium sized mammals.
– Specify a chip fabrication process supporting 1 million neurons per square centimeter and ten billion synapses per square centimeter.

FY 2012 Plans:
– Design and simulate in software a complete neural system of ~10 billion synapses and ~1 million neurons performing cognitive tasks in a virtual environment comparable to those routinely tested in mice.
– Design and verify a hardware neural system of ~10 billion synapses and ~1 million neurons.
– Demonstrate a chip fabrication process and development plan supporting ~10 billion synapses per square centimeter and ~1 million neurons per square centimeter.
– Refine design tools and techniques by codifying design rules and component properties and matching them to fabrication and
simulation capabilities.
– Demonstrate a virtual environment supporting visual perception, decision and planning, and navigation environments fully integrated with software or hardware neural systems enabling the testing, training, and evaluation of these neural systems.
– Expand the feature set of the virtual environment to include auditory perception and proprioception.
– Introduce modalities of competition within the virtual environment to further tailor the evolution of the neural systems

If you liked this article, please give it a quick review on ycombinator or StumbleUpon. Thanks