



	CM-1K Neural Network Chip

	CM-IR2K Image Recognition Board

	CM-SK1K Starter Kit

	CM-PM1K Prototyping Module

	CM-EB2K Evaluation Base Board

	CM-EMB Embedded Module

	CM-V1KU Vision Sensor Board

	Contact C4AV

	Forums

	Home

CM-1K Neural Network Chip

CM-1K Neural Network Chip is a powerful neural network chip featuring 1024 neurons working in parallel and a parallel bus which allows to increase the network size by cascading multiple chips. It is an ideal companion chip for smart sensors and cameras and can classify patterns at high speed while coping with ill-defined data, the detection of unknown events, and adaptivity to changes of contexts and working conditions, etc.

In addition to its parallel neural network, CM-1K integrates a built-in recognition engine which can receive vector data directly from a sensor and broadcast it to the neurons in real-time. Upon receipt of the complete vector, the category of the firing neuron with the closest match is transmitted to the output bus. In the case of a monochrome video sensor, CogniMem offers a proprietary signature extraction from 2D video to 1D vector. The recognition engine can operate at sensor speed (up to 27 Mhz). The usage of the high-speed recognition engine requires that a knowledge be previously loaded into the neurons.

Purchase The
CM-1K
Neural Network Chip »

View CogniMem SDK - Control or simulation of the CogniMem_1K chip With optional functions for the CogniMem evaluation boards.

Applications:
Image Recognition	Signal Recognition	Pattern Search
Face recognition Fingerprint identification Target tracking Target identification Factory inspection Person monitoring OCR-Gaze tracking Human System interface Biological imaging (microarrays) Wild intelligence sensor Kinematics monitoring Smart airbag Much more…	Speech recognition Voice identification Radar identification EKG monitoring EEG monitoring Sonar identification Spectrum recognition Flight analysis Vibration monitoring Much more…	Cryptography Genomics BioInformatics Unstructured data mining CRC Much more…

A Parallel Neural Network
A neuron is a reactive memory which can autonomously evaluate the distance between an incoming vector and a reference vector stored in its memory. If this distance falls within its current influence field, it returns a positive classification.

The true significance of a neuron is its arrangement into a parallel network which allows to decode the collective response of all the neurons in a constant amount of time. This response can be a list of categories and distances automatically sorted per decreasing confidence level. An empty list means that the vector is not recognized. More Details. Prototype=256 bytesContexts= 127 Distance norm= L1 or Lsup Category= 32768 Neurons= 1024

Unique Recognition Capabilities
Constant recognition time after vector broadcast to the neurons, independent from the number of neurons in use Global response status: Positively identified, Identified with uncertainty or Unknown Detailed response of the firing neurons: Distance value between input and prototype, Category value of the prototype. This data is retrieved per firing neuron per increasing distance value (i.e. decreasing confidence level) Recognition under multiple independent contexts for data fusion and hypothesis generation- Anomaly detection and predictive maintenance through the detection of an Unknown classification status
More Details… Classification status after 1 clock cycle Category and distance inquiry = 36 clock cycles Distance inquiry = 16 clock cycles Contexts=127 Recognition cycle= 10 usec for 1 256-byte vector

Automatic Model Generator and Adaptive Learning
Transfer human expertise by teaching examples of vector data. The neural network builds the corresponding knowledge on its own. Add more training at any time to expand or complete a knowledge base. The neural network will adapt to fit any example adding novelty to an existing knowledge. The throughput and accuracy of the knowledge can be tuned to produce a recognition engine with conservative or moderate behavior.
Learn by examples (supervised or unsupervised) Automatic model generator Map decision spaces by aggregate instead of hyper planes Cope with non-linear, convex, disjoints and embedded categories Modulation of throughput versus accuracy Multiple space generation using different context for data fusion and hypothesis generation Novelty detection Save and restore the contents of the neurons.
More Details… Nearest neighbor model Compound Classifier Space dimension=256 Number of spaces or contexts= 127 Basic shape entity= Square or Diamond

Optional High-Speed Recognition
In addition to its parallel neural network, CogniMem integrates a built-in recognition engine which can receive vector data directly from a sensor and broadcast it to the neurons in real-time. Upon receipt of the complete vector, the category of the firing neuron with the closest match is transmitted to the output bus. In the case of a monochrome video sensor, CogniMem offers a proprietary signature extraction from 2D video to 1D vector. The recognition engine can operate at sensor speed (up to 27 Mhz). The usage of the high-speed recognition engine requires that a knowledge be previously loaded into the neurons.