Congimem CM1K - Neural Network Chip


11 posts
by DLPurcell » Mon Jul 09, 2012 7:14 pm
Hi there,

I'm patiently waiting for my RPi to arrive and I'd like to share one of my project ideas. I'm thinking of hooking up one of these using the GPIO ports. It's a hardware neural network chip.

I'd eventually like to have a wearable computer that I can teach to recognise images and sound. But the first stage is going to be talking to the chip using GPIO.

Does this interest anyone? Any thoughts? Any foreseeable problems?

Dan
Posts: 3
Joined: Mon Jul 09, 2012 6:26 pm
by rasbeer » Tue Jul 10, 2012 11:16 am
It's a nifty idea - how much do those chips actually cost?

People have recently got openCV working on Pis (link), but maybe that wouldn't be fast enough for what you've got planned?
Posts: 242
Joined: Wed Mar 07, 2012 8:35 am
by bobc » Tue Jul 10, 2012 11:37 am
From the presser CogniMem™ Launches CogniBlox™

The CogniBlox system is composed of four CM1K chips or a total of 4096 cognitive memory processing elements per board in a trainable 3-layer network, each having 256 programmable 1-byte connections to the input. Systems of 1 million elements can be configured allowing for 256 million connections every 10 microseconds with a typical power consumption of 500 watts and 0.13 petaops1 of performance. The Cogniblox system is supported on Windows and Linux platforms using .NET and Java based tools for managing the simple commands for training and recognition tasks of the array. Hardware support for dynamic re-configurability and complete architectural flexibility for X, Y & Z scaling is also inherent in the CogniBlox architecture. Prices start at $3,000 for a single board CogniBlox, with volume discounts for the 1 million system (250 CogniBlox boards).


Not quite "cheap as chips"...
Posts: 86
Joined: Fri Apr 06, 2012 8:01 am
by rasbeer » Tue Jul 10, 2012 11:41 am
Aren't neural networks 'massively parallel'? $3000 is quite a few Pis...
Posts: 242
Joined: Wed Mar 07, 2012 8:35 am
by DLPurcell » Tue Jul 10, 2012 1:03 pm
Actually, a breadboard ready version of the chip only costs $149

This technology is definitely within reach of us mere mortals.

The CogniBlox is an array of 4 of these chips. It comes with the SDK which is what bumps up the price.

OpenCV would be way to slow on the Pi unfortunately.
Posts: 3
Joined: Mon Jul 09, 2012 6:26 pm
by gritz » Tue Jul 10, 2012 2:00 pm
$149 for proto board with chip.
Posts: 449
Joined: Sat Jan 28, 2012 2:33 am
by bobc » Tue Jul 10, 2012 3:20 pm
Ah, that's more affordable. Could be some really interesting applications. I would like something to work out those obscure CAPTCHAs for me.

Anyone seen the Human Brain Project? Here comes the Singularity!
Posts: 86
Joined: Fri Apr 06, 2012 8:01 am
by gritz » Tue Jul 10, 2012 4:21 pm
DLPurcell beat me to it!

I've always beem fascinated by neural networks. In 1991 I would have bet a weeks beer money that by now hardware neural nets would be everywhere and digital would be relegated to doing very_hard_sums. Simulating them in software on pcs just seems like missing the point to me! :)
Posts: 449
Joined: Sat Jan 28, 2012 2:33 am
by DLPurcell » Tue Jul 10, 2012 5:06 pm
gritz .. Yeah, simulating them is very slow. It gets exponentially slower the bigger the net too. But these chips can be linked together. You can clock one on it's own @27MHz, but when you link more than one, it's a one off cost of half the speed. And as far as I understand, you can just keep daisy chaining more and more together.

bobc .. Have a look at some of the demo videos. I can't see these chips having any trouble reading current capchas .. Looking into the Human Brain Project now, very interesting.

On a side note, here is a little neural net program I wrote a while ago (java, runs in browser)
Posts: 3
Joined: Mon Jul 09, 2012 6:26 pm
by xvision » Tue Jul 10, 2012 8:32 pm
If you are interested in high performance neural networks on the RPi I would recommend you look at GPU programming using GLSL ES..
Posts: 4
Joined: Mon Jul 09, 2012 10:58 am
by Kerron » Thu Sep 19, 2013 12:11 pm
What type of Neural Network chip are going to use?
Posts: 1
Joined: Thu Sep 19, 2013 11:53 am