Singular Computing’s S1 Chip — The Next Big Thing
A new approach to computation offers comparable precision with twice the speed and lesser power consumption.
Since the story is there, hence it must be told. Once, a man arrived at a river. He needed to cross it, but there was no bridge; so he needed to know the width of the river.
Had he been an ignorant shepherd with his innocent flock of sheep, he would have turned around for good and wondered no more about this river. But this was Napoleon Bonaparte -the Little Shepherd of the Giant Europe.
And the curiosity of great men must necessarily be quenched. An engineer was instantly summoned from the nearby town. It is not every day that town engineers have an emperor waiting for them — the baffling nature of the appointment confounded the engineer’s brain and legs. He came late, and without his instruments.
Unlike the general populace, Napoleon harboured a great dislike for negligence. He ordered his subordinates to see that the engineer measure the width of the river by the end of the day, otherwise he must be stripped of his post and pension.
Times of great peril have amazed with the greatest performances of wit. The engineer was no exception. He composed himself calmly and took off his hat, tilting his hat on his eyes, he moved a certain distance parallel to the river. He stopped and wrote something on the riverbank with a stone. And did that again from another spot. He had calculated the approximate width with the rule of triangles! While not exactly accurate, it was nonetheless and innovative way to find a solution; a juggad.
The moral — sometimes approximate jugaads can save your life.
The same moral applies on the area of computational chips. The world of computation is founded on the absolute correctness of even the tenth decimal — something the world can do without. This is what Joseph Bates, the CEO of Singular Computing LLC, thinks.
Backed by technological reservoirs like DARPA, this thinking has been instrumental in the progression of Singular Computing’s chip-S1. According to Bates, a 99% correctness of the S1 chip requires a hardware of a nature that consumes very little power and delivers results at a much, much faster rate.
The novelty of S1 chip resides in the fact that it has a core arithmetic unit that does “floating point operations“ (add, subtract, divide, multiply, and square root) in a single cycle with the help of a certain algorithm that forms an APE (approximate processing element). These APEs communicate with each other over a skeletal parallel grid.
Compared to a normal computing chip, S1 produces comparable results using 1/100th the silicon area and in half the processing time that other chips need.
According to the estimation of Bates, the current technology could enable the cost-effective integration of several hundred thousand APEs on a single chip, alongside an ARM or other host processor.
On another evaluative test performed by the Office of Naval Research with Singular Computing in partnership with Charles River Analytics, it was concluded that APE based hardware algorithm can deliver comparable quality results at 89x the frame rate of the software only scheme, all the while consuming 72x less power!
All these results show that a CPU managing low-precision “workers” can yield high-precision CPU-like results, with significant advantages in size, weight, power and cost.
All these developments might usher in a new era of approximate computation — most of which shall be used in making the size of the hardware smaller and getting cost effective data.
It’s uses in industries like smartphones and surveillance technologies is anticipated, and so is in artificial intelligence. Bates’ chip, with the help of his research and DARPA might be setting up some new parameters in the tech schema and may even change consumer technology at it’s bedrock… processor performance.
Originally published at Chip-Monks.
