CogniMem: The Embedded Vision Alliance's Latest Mem(ber)
Those of you who have visited the Embedded Vision Alliance's member page in the past few days may already have noticed what I'm announcing here; that the Alliance has a new member! It's CogniMem Technologies, a company who I previously mentioned back in early January by virtue of their presence in Freescale Semiconductor's product suite at the Consumer Electronics Show. That particular demo showcased gaze tracking (i.e. eye-controlled user interfaces), but CogniMem has broader embedded vision aspirations, as the company's description on the EVA website suggests:
CogniMem is leading the industry with cognitive sensing and computing. You train the chip by showing it image data (still or video, or any digital data) wherein it automatically generates models of the information that is seen. For image recognition, it provides orders-of-magnitude of performance/watt advantage over leading industry CPUs. The technology provides a fuzzy match or a distance vector in a constant 10 µsec, regardless of the number of vectors that are being searched against (ex: 1/1 million = 10 µsec). It is a non-linear classifier component that natively implements k-NN and Radial Basis Functions in hardware via a memory-based processing approach (machine learning neural network), ideal for gesture recognition, visual target tracking, face recognition, anomaly detection, video analytics, robotic vision, visual inspection and more.
EVA founder Jeff Bier and I both first became acquainted with CogniMem during last summer's Freescale Technology Forum, held a few weeks before I officially joined the Embedded Vision Alliance. But my indirect association with the company goes back many years prior. CogniMem's co-founder, president and CEO Bruce McCormick was, between 1991 and 1996, my boss as the marketing manager for Intel's flash memory division. It's a small world after all...
For additional information on the company's core technology and products based on it, McCormick offers up two specific pieces of collateral:
- A paper, "A hardware/software co-design model for face recognition using Cognimem Neural Network chip," published at the November 2011 International Conference on Image Information Processing (ICIIP), and
- A technical reference manual, "Target Tracking with the Zero Instruction Set Computer: Evaluation of the Computational Capacity of the ZISC in Target Tracking"
And, of course, you can also visit the company's website, which contains an abundance of information on the CM1K chip, CogniBlox evaluation board and V1KU evaluation kit (the latter co-developed with fellow EVA member Freescale), and both online and face-to-face training opportunities.