|
|
| |
CM-EB2K Evaluation Base Board |
| |
The CogniMem Evaluation Base Board (CM-EB) offers developers and OEMs a comprehensive platform to evaluate the CogniMem neural network technology for the real-time recognition of data coming from sensors, instruments, or else. Typical applications include machine vision, face recognition, voice and signal recognition, but also data mining. The board features 2 CogniMem chips, each with 1024 neurons, an Actel FPGA accessible to programmers and a digital input bus for easy connectivity to external sensors. User I/O lines include an I2C serial bus, an RS232 bus and 8 uncommitted general purpose I/Os brought to header pins. |
 |
|
| |
The CogniMem neural network implements two powerful non-linear classifiers (RCE and KNN) in a natively parallel architecture. The tremendous benefit of this architecture is a recognition cycle which remains under 11 us whether the entire network is composed of 1, 2 or more chips. Brute computational power is equivalent to 80 gig operations/second @ 27 MHz for one chip, twice as many for two chips, etc. The CogniMem neurons build their knowledge by learning example vectors and their associated category. This can be done in real-time or a pre-existing knowledge can be loaded in advance from file or Flash memory.
|
| Specifications |
| |
| |
Neural network |
|
Classify pattern vectors of up to 256bytes |
|
Up to 32768 categories |
|
Classification status in 1 clock cycle |
|
Category readout in 36 clock cycles per |
| |
up firing neuron from smallest distance |
| |
and (equal to 3 microsec at 27 Mhz) |
|
Radial Basis Function or K-Nearest
|
| |
neighbor classifier |
|
Automatic model generator |
|
Expandable through neuron expansion |
| |
modules |
I/O buses |
|
Miniature USB Hi Speed (480 Mbps) |
|
I2C serial interface (100-400 kbit) |
|
Serial output (115,200 baud) |
|
8 parallel outputs (LVTTL 16mA) |
|
Save project to Flash memory |
|
|
High-speed recognition engine |
|
|
|
input clock signal (up to 27 Mhz) and
|
| |
vector valid signal |
|
Output category with the best match 3 us
|
| |
after receipt of the last vector data |
|
line valid input signal |
|
region of interest as small as 16x16
|
| |
pixels and as large as a full frame |
|
Built-in signature extraction |
|
Connector compatible with Micron sensor
|
| |
demo board |
FPGA Program |
|
Optional core modules (input and output |
| |
data conditioning, signature extraction, |
| |
decision logic, etc. |
Mechanical and Electrical |
|
3.3v @ <250 mA |
|
55กม66mm |
|
|
| |
Ordering information |
| |
|
CM-EB, includes: |
| |
|
One CM-EMB2K module, |
CogniMem development library |
Easy Video Trainer software |
|
Optional stackable CM-EMB2K modules with 1024 or 2048 neurons |
| For more information, you can download some documentations from here. |
|
|
|
