datasheets.com EBN.com EDN.com EETimes.com Embedded.com PlanetAnalog.com TechOnline.com  
Events
UBM Tech
UBM Tech

News & Analysis

Comment


Frank Eory

6/16/2011 6:58 PM EDT

It seems like FEC decoding is the the most obvious application for this ...

More...



ece-professor

6/16/2011 1:34 PM EDT

Analog decoders for codes like Turbo and LDPC codes have been studied for over ...

More...

ADI buys Lyric – probability processing specialist

Junko Yoshida

6/13/2011 5:01 PM EDT

NEW YORK – Analog Devices, Inc. (ADI) disclosed Monday (June 13th) that it has acquired Lyric Semiconductor, a privately-held start-up based in Cambridge, Mass.

Lyric, a DARPA and venture-funded MIT spin-out founded in 2006, is known for unique innovations in probability processing.

At a time when a significant proportion of computing and embedded applications depend on the science of guessing – or the math of probability— software algorithms based on probability are everywhere, explained Sam Fuller, CTO at ADI and vice president of R&D. But the Lyric team is believed to be the first to develop hardware circuit architecture optimized for probability processing. “Lyric is definitely ahead of the pack,” said Fuller.

Fuller described the deal with Lyric “a big acquisition on a big idea,” which brings in new innovations to signal processing. The entire Lyric team consisting of 26 people, including co-founders Ben Vigoda and Dave Reynolds, are now ADI employees. ADI, however, declined to disclose how much the company spent on the Lyric acquisition. The deal was completed on June 9th.

How real is the technology?

Lyric got the engineering community’s attention when it showed off its first commercial application — Lyric Error Correction for flash memory. Lyric’s probability processing technology, now up and running at Lyric’s labs in Cambridge, has enabled orders-of-magnitude improvements in processor efficiency.

Take, for example, data stored in a 30nm flash memory. According to Lyric, data read out by the flash controller today has an error rate of about 1 bit wrong in every 1000 stored. By using Lyric’s probability gate circuits in a highly parallel architecture, Lyric demonstrated that its technology can improve the error rate to “one in every trillion, or 1,000 trillion,” according to Vigoda. “Our technology converts analog information stored in flash memory into correct digital representations,” he said.

As a result, “performance wins of 30X and over 10X in power consumption” are possible, according to Lyric. Fuller noted, “A clear value to our customers is power efficiency. As it improves an error rate, our customers can also use more sophisticated architecture in their products.”

Lyric Semiconductor won this year’s EE Times ACE Award in the category of the most promising new technology.


What does this mean to ADI?

While Lyric clearly has a unique approach in probability processing, it’s up to Fuller’s team to figure out how truly relevant and broadly applicable Lyric’s technology is to ADI’s product lineup.

While licensing Lyric’s technology to other chip companies (such as flash memory vendors) is a possibility, Fuller made it clear: “We did not buy Lyric for IP play. We acquired to innovate our products further.”

Fuller noted that his team was first introduced to Lyric six to eight months ago by Ray Stata, who was a co-founder of ADI and is now a lead partner of Stata Venture partners, currently serving as Lyric’s chairman of the board.

ADI first collaboration with Lyric is the use of Lyric’s technology in the communication infrastructure space, such as base stations, explained Fuller. The project, still in development, is focused on filling in gaps ADI faces in the signal chain in telecommunications – such as signal conditioning and prepping – typically sitting next to ADI’s high performance converters, for example.

Fuller was careful not to pre-announce when ADI will roll out the first product using Lyric’s technology, noting that the introduction of any new technologies involves certain risks. But as he has gotten a deeper understanding of Lyric, he said he has become confident that ADI will be able to take full advantage of its ideas. “There are potentially much broader implications. It’s our responsibility to define two or three places to focus [in applying Lyric’s innovations to] and bring them to the market,” he said.

Potential areas where Lyric’s technology could shine in ADI’s signal processing products include: automotive, communication infrastructure, health care and consumer (including gesture control), according to Fuller.

While the company is known for many technology innovations in recent years such as MEMs, gyroscopes and accelerometers, ADI hasn’t actively pursued many merger and acquisition opportunities. “Typically, we’ve done many innovations internally and organically,” said Fuller.

The last acquisition Fuller can remember his R&D team did was in 1999 when it bought Edinburgh Portable Compilers of Scotland. Meanwhile, the last acquisition ADI corporate pulled off was in 2006, when it acquired privately held AudioAsics A/S, a Danish company with expertise in high-performance, low-power microphone and audio signal conditioning technology.


Does the deal with Lyric mean a change of directions for ADI?


Yes and no. Clearly, with Lyric, ADI is willing to take some risks. Although ADI has often done collaborative work with university contracts, etc, Fuller said, “As our business spread out, I think we have become more open to bring innovations from outside.”

Fuller pointed out that the beauty of Lyric’s technology is that it uses the same standard CMOS technology – such as 180-nm or 65-nm process technology – used in current mixed signal products. “The same packaging technology can be applied,” he added. ADI is confident that the company will bring to Lyric’s innovations “our global sales and distribution networks and world-wide manufacturing capabilities,” Fuller summed up.




junko.yoshida

6/13/2011 5:31 PM EDT

Has anyone out there seen any hardware architecture "optimized for probability processing"? How useful do you think will it be -- especially in which of your products you are working on?

Sign in to Reply



agk

6/14/2011 9:45 AM EDT

Highly parallel architecture is advantageous in all the applications because they are fast. So more of the probability mathematics can be applied and tested in real time for improved performance.Lyric worked on this for 5 years and won.

Sign in to Reply



GREAT-Terry

6/15/2011 10:53 AM EDT

Maybe ADI can build lower power and yet high performance DSP that is really sexy. TI I think is their target to beat!

Sign in to Reply



Gil Russell

6/15/2011 4:27 PM EDT

Good report Junko, thanks.
Lyric has apparently met with difficulty in finding design wins for their technology. This is not uncommon for something that is "completely different" from the status quo.
Applying it to NAND-Flash memory for error correction using Low Density Parity Codes[LDPC]is an appealing mass market application. Why Lyric was not successful here would be an interesting story by itself. It will be interesting to follow where ADI applies Lyric's technology - that is if that information is "allowed" to be made public...,

Sign in to Reply



ece-professor

6/15/2011 8:55 PM EDT

Lyric's solution has no any advantages over (if is not much worse than) existing pure digital-based LDPC codec being used by companies like LSI and Marvell.

Sign in to Reply



junko.yoshida

6/16/2011 10:03 AM EDT

Good point, Gil. Why wouldn't have NAND-Flash guys gone all over Lyrics before ADI snatched them up? I don't know. But definitely something to pursue.

Sign in to Reply



gsdg90

6/16/2011 10:00 AM EDT

Error Correction of waveforms is conventionally performed by digitizing that waveform (using an ADC) then applying a 'soft decision' error correction algorithm to the quantized waveform levels.
It is a cute trick to perform the whole error correction algorithm using an 'analog computer' that relies upon the I-V relationships of the transistors. The big question is whether or not this offers any kind of advantage to the 'digital computer' approach, particularly on sub-micron CMOS.

Sign in to Reply



ece-professor

6/16/2011 1:34 PM EDT

Analog decoders for codes like Turbo and LDPC codes have been studied for over 10 years in academia and there were some impressive silicon demonstrations, and even few startups were founded but all of them failed eventually.

Sign in to Reply



Frank Eory

6/16/2011 6:58 PM EDT

It seems like FEC decoding is the the most obvious application for this technology, and they're using it for that already. But I wonder what this could do for machine vision?

Sign in to Reply



Please sign in to post comment

Navigate to related information

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)