Micron's Automata Exploits Parallelism to Solve Big Data Problems

eetimes.com

38 points·hershel··7 comments

For more details about this processor:

http://www.micron.com/~/media/Documents/Products/Other%20Doc...

Peformance:

It uses DRAM process.for a snort test(intrusion detection) this (relatively small) chip compares to 9 large FPGA chips.it takes 1/9 of the power. They have 8 chips on a dimm module.

Question: what are some interesting applications for this ? and how do they map to finite automata ?

This type of processor implements nondeterministic finite state machines². Effectively it allows you to perform regular expression matching, very fast and on large data sets. They implemented this in a very clever way exploiting the fact that DRAM access is extremely parallel by nature. It is useful to solve data mining problems, such as those arising in bioinformatics.

It is quite likely that there are also certain government agencies among the early customers...

²http://en.wikipedia.org/wiki/Nondeterministic_finite_automat...

Btw, a pretty cool thing about their implementation is that they use a standard DDR3 interface which allows the realization of these machines in a standard DIMM form factor. This means that they could be added to almost any existing server.

Fascinating. Can we add a stack?

I believe you are thinking of a different computing paradigm.

I am thinking about what it takes to turn a NFA into a push-down automata.

I wonder if this would be good for features in computer vision too.

Reply on news.ycombinator.com