The holy grail of computing—a single chip for everything—is the stuff of science fiction. But a small start-up incubated in the Indian Institute of Science’s (IISc) Society for Innovation and Development is working on a technology that is as good as it gets in the real world.
Morphing Machines Pvt. Ltd, founded by S.K. Nandy, a professor at IISc’s Supercomputer Education and Research Centre in Bangalore, has designed its first market-ready application in the emerging field of reconfigurable computing.
The concept of reconfigurable computing has been around since the 1960s, but efforts by countries such as Japan and the US to create applications in automobile engineering, avionics, multimedia, mobile technologies and even personal computing have not been successful.
Different approach: Morphing Machines’ founder S.K. Nandy (in blue shirt) in the REDEFINE lab at IISc, Bangalore. Hemant Mishra/Mint
Consider this. A firm wants to make a chip for set-top boxes. Typically, it takes eight-nine months to develop one from scratch. Say six months down the line, the firm wants to manufacture a multimedia chip for a smartphone. It would be staring at another eight-month development cycle. Enter reconfigurable computing. “Think of it as technology that can dramatically cut down time and investment in new chip development by allowing the same hardware to be used for multiple purposes,” says Ganesh Ramamoorthy, principal research analyst (semiconductor IP and design industry) at Gartner Inc.
Simply put, a new circuit can be loaded on the same chip, in effect reconfiguring it for a different purpose. With this, a chip manufacturer can simply license a reconfigurable multimedia chip design from a vendor such as Morphing Machines. With some minor tweaking to the circuits, it can get both a set-top box chip and a smart phone multimedia chip to the market in four-five months.
Decoding the chip
That, then, is the technology in which firms such as Intel Corp., International Business Machines Corp. (IBM), Akya Ltd, Celoxica Holdings Plc. and Panasonic Corp. have been investing for years. Success has been elusive.
When computer scientist Gerald Estrin first proposed the concept in the 1960s, implementation was tough because the idea was far ahead of the technology existing then. In the 2000s, even with better silicon technology, the heavy investments required to make reconfigurable computing work made it unviable.
So, while Morphing Machines says the worldwide market for reconfigurable computing can be pegged at about $19 billion (around Rs 85,000 crore), Ramamoorthy says penetration is less than 1%.
Developing a reconfigurable chip is an expensive, resource-heavy exercise. A single chip requires a large number of engineers working on it. The higher the customization or reconfigurability, the more complex the design, with interoperability, power management, chip interconnections and electronic interference becoming problems.
This is where Morphing Machines seems to have made some headway. Any encryption program, whether in mobile phones such as the BlackBerry or email, uses a chip that can encode the information at a particular strength. A BlackBerry would use a 128-bit AES encryption key, while really critical applications such as government communications would require 256-bit encryption. A single chip can only encrypt at a single strength.
With Morphing’s REDEFINE platform, a single chip can be created to encrypt at any strength depending on the user’s needs, without compromising on speed. This helps create highly flexible encryption solutions that are much harder to attack. The application is now in trials with clients in the government sector.
REDEFINE has four patents, and research on it has been published in the peer-reviewed journal, Transactions of Embedded Computer Systems of the Association for Computing Machinery. Additionally, the University of Tokyo and the University of Leiden, the Netherlands, have adopted it as a research vehicle for their own work.
So, how is Morphing’s approach different, and is it likely to see greater success? Says Nandy, “Our approach is based on coarse-grained abstractions, as opposed to many efforts so far that have been fine-grain.” This means Morphing does not tackle the problem at the basic level of machine-language, but at a higher level.
“Another common approach uses field programmable gate arrays. These are circuits that allow a client to reconfigure them the way he wants. Again, this method takes longer and is more complicated than ours,” says Chandan Haldar, chief strategy officer at the three-year-old firm.
“Secondly, we have defined our problems more narrowly by focusing on a set of niche applications like cryptography and cognitive radio. The focused definition makes our solutions highly effective in each of these niches,” he says.
Morphing Machines is also developing applications in areas such as video processing, cognitive radio and transcoding. On video processing applications, Haldar says, “What we receive today in the name of hi-definition TV through our set-top boxes is often not HDTV. Real HDTV is at a resolution of 1,080 pixels, but your set-top box may be incapable of rendering it, despite what the TV station transmits. Morphing Machine’s chip can rest on the set-top box and reprogramme according to the resolution of the broadcast.”
As of today, despite its potential, reconfigurable computing is at least three-five years away from peak adoption, says Ramamoorthy. “A lack of design tools such as embedded programmable logic blocks and the high cost of development is a cause for concern in this field,” he adds. While most large chip makers have research programmes in this area, there are few independent firms, especially in India, who are investing in the technology.
When contacted, IBM and Intel acknowledged that they had programs in the area, but refused to comment further on the same.