The CogniMem^™ CM1K Chip is the first ASIC version of the CogniMem neural network product line optimized for Cognitive Computing. It features 1024 neurons working in parallel implementing two reknown non-linear classifiers. It can recognize patterns at high speed while coping with ill-defined data, unknown events and changes of contexts and working conditions. Furthermore, CogniMem CM1K Chips can be daisy-chained to other CogniMem CM1K Chips to increase the network size. Its low pin-count and low power consumption make it an ideal companion chip for smart sensors and cameras. Its parallel architecture make it an ideal candidate for large data mining systems.

This plot of the CM1K ASIC shows that it is a set of 1024 identical cells, or neurons. They are connected in parallel through a bus of 33 lines. At 27Mhz, the CM1K is 90 times faster than a DSP at 300Mhz to match one pattern of 256 bytes among 1024.

 

 

 

 

PRODUCT SPECIFICATIONS

 

• 1024 neurons in parallel on a chip
• Learn and recognize patterns up to 256 bytes
• Classify patterns up to 32,768 categories
• Automatic adaptive model generator
• Choice of Radial Basis Function (RBF) or K-Nearest Neighbor (KNN) classifiers
• Recognition time independant of the number of neurons
• Possible segmentation of the neurons up to 128 contexts (per feature, per sensor, etc.)
• Save and restore knowledge
• Simple RTL protocol with less than 2 dozen registers
• Low cost, small footprint, low power consumption (.5w)
• Ability to interconnect multiple chips with a simple bus of 28 lines