CogniMem Technologies Inc. has developed a greater than 40,000-neuron, scalable cognitive memory computing system based on CogniBlox, a memory-based parallel processing capability that architecturally implements how the human brain processes data. As a result, computing can scale on hardware without any additional software adjustments. CogniMem introduced and displayed the system here at the 2012 Supercomputing Conference.
With architectural re-configurability and local Magnetoresistive Random-Access Memory (MRAM), the CogniBlox system can scale to very large arrays of cognitive memory and configurations. The result is a platform for deployment of real artificial intelligence with exceptional speed performance and low power consumption. The greater than 40,000 neuron cognitive memory system consists of 10 CogniBlox boards, offering a large bank of cognitive memories with no impact on operation and enabling a path to exascale computing for a wide range of data-mining applications.
CogniBlox architecture is based on alignment of multiple CM1K (1,024 memory processors each) components. The technology allows for constant parallel matching of vectors in 10 microseconds, regardless of the number of vectors being compared at the same time. The CM1K chip is designed for determining closest vectors in video searching, real-time surveillance and analytics, data mining, genomics, fingerprint matching, hyperspectral image analysis, financial services, weather forecasting and a wide range of scientific computational tasks.
Prices start at $3,000 for a single board CogniBlox (4,096 (or 4K) Cognitive memories, with volume discounts for the larger multi-board Cogniblox configurations.
CogniMem is demonstrating its 40,000-neuron cognitive memory system Nov 13-15 at Booth # 1534 at the SC12 conference in Salt Lake City..