To Lyric Semiconductor founder and CEO Ben Vigoda, the world of computing is moving away from the binary 1s and 0s and toward probabilities.
The yes/no nature of binary computing is fine for such tasks as operating systems, databases and spreadsheets, Vigoda said in an interview with eWEEK. But many applications that are becoming ubiquitous in today’s world-from fraud detection and financial modeling to genome sequencing and search-are not looking for the yes/no answers, but rather the best guess answer from among any number of possibilities.
Digital processors, such as those developed by Intel and Advanced Micro Devices, can do these probability calculations, but in an extremely inefficient manner, Vigoda said. As a result, such calculations call for huge amounts of processing power, space and money. With Lyric Semiconductor, Vigoda and other company officials are looking to create a new processing environment that calculates such probabilities much more quickly and efficiently.
Lyric, an MIT-spinoff that is funded through DARPA (Defense Advanced Research Projects Agency) and venture capital firms, came out of stealth mode Aug. 17 with the first of its commercial products based on its new computer circuitry, LEC (Lyric Error Correction) for Flash memory, a technology that can be found in everything from smartphones and cameras to servers and storage devices.
The 65-nanometer LEC chip, manufactured by the Taiwan Semiconductor Manufacturing Co., will help alleviate the ECC bottleneck found in Flash memory, Vigoda said. Currently, one in every 1,000 bits of storage in Flash memory comes out wrong when the memory is read, he said. In addition, as density is increased in Flash memory, so are error rates.
As this happens, error correctors have become larger, increasingly complex and more expensive, according to Vigoda. Lyric’s LEC chip is 30 times smaller than those Flash memory ECCs, is 12 times more power efficient, and is faster than traditional chips.
Lyrics Flash error correction chip essentially will serve two purposes. The first will be the company’s first commercial product, and one that could save billions of dollars in the industry. The second is as a proving point of the new processing circuitry that Lyric is developing.
Lyric’s technology was developed from the work Vigoda did more than a year ago for this doctoral thesis at MIT. He saw the need for a new way of processing, one that worked out probabilities.
For example, if a person’s credit card is used to buy an item, the computer must determine the probably that the charge was fraudulent. The computers for Netflix or Amazon, based on data from a person’s previous film rentals or book purchases and what they’re clicking on at that moment, must calculate probabilities when determining what to suggest for future rentals and purchases.
And whereas such calculations may take hundreds of traditional processors a long time to calculate, Lyric would only need a handful of its processors-which are designed to handle multiple probability calculations at one time-over a much shorter amount of time. Whereas chips from Intel and AMD use 0s and 1s, Lyric’s processors also calculate numbers between 0 and 1.
“We’re accomplishing with a few transistors what traditionally would need to be accomplished with hundreds or thousands of transistors,” said Mira Wilczek, director of business development for Lyric.
That will lead to significant savings in everything from power and size to costs, Wilczek said.
In addition to the LEC, Lyric also is working on a general-purpose chip based on the company’s probability processing platform. The GP5, which will begin sampling in 2013, will be designed to calculate probabilities for everything from Web searches to genome sequencing, and will offer more than 1,000 times the performance of current systems based on x86 processors from the likes of Intel and AMD, Vigoda said.
“The number of genomes people are sequencing are doubling every year, faster than Moore’s Law can keep up with,” he said.
Code for the GP5 will be written in Lyric’s own PSBL (Probability Synthesizes to Bayesian Logic) programming language, created to work with probability-based calculations. The company is planning to use the combination of its processors and programming language to create new systems that will challenge x86-based servers, Vigoda said.
Lyric is working with DARPA to develop the GP5 and PSBL, and he is hoping to start catching the eye of businesses.
“We’re looking for the early adopters,” Vigoda said. “We’re already [working with the Department of Defense]. The next step is finding early adopters on the commercial side.”
He said he understands the challenges of trying to break into an area of the industry dominated by Intel, but believes that once Lyric’s technology gets out into the market, people will start gravitating toward it and it will catch on.
“We’re well aware that the ground is littered with the corpses of [companies that have made such attempts],” Vigoda said.
The most notable of the recent attempts was made by Transmeta, which created low-power processors designed to challenge Intel and AMD. Transmeta has since gone out of business, but not before turning the other two vendors onto the demand for greater energy efficiency in processors.
Now other vendors are jumping into the mix, looking to take advantage of the growing demand for low power and the rise of such computing models as virtualization and the cloud to introduce new processor and server designs.
Tilera, which already has 36- and 64-core processors on the market, is planning to roll out a chip with 100 cores next year, and a 200-core processor in 2013. It also announced an ODM (original design manufacture) contract with Quanta Computer to develop a system that can put 512 computing cores into a single rack.
Meanwhile, SeaMicro is planning to roll out a system that can scale to 512 Intel Atom processors, with the goal of giving OEMs and businesses a less expensive alternative that offers the same performance as traditional Intel- and AMD-based servers.