Starting chip companies was never for those short on capital. From the moment Art Rock backed two of Fairchild Semiconductor’s “traitorous eight” with $2.5 million when they split off to found Intel, the up-front costs have been soaring. First there was the R&D. Then the logic designers and the chip designers had to figure out how to build the chips. And then there were the fabs, which today cost in excess of $1 billion each to build. Even with outsourcing manufacturing, it costs an average of $25 million to $50 million to start a new chip company these days. Imagine if you could slash that cost by 75%.
Don’t believe it? Just talk to Suhas Patil, founder of Cirrus Logic and one of the fathers (along with Gordie Campbell) of the fabless semiconductor concept.
With Cirrus, Suhas was able to build a company around a very simple concept: Build technology that permits even those not trained in silicon technologies to design integrated circuits in six months. System designers could design complex chips and bring them to market without the need to become experts in transistor or circuit design. “Early on in Cirrus Logic I had figured out that one did not have to build a fab to be a semiconductor company,” Suhas says. “And if Moore’s Law was going to continue to hold true, someone needed to develop design automation so that the logic designers themselves could complete chip designs.”
Thanks to companies like Cirrus Logic and Chips and Technologies (founded by Gordie), the fabless semiconductor industry was born. Now, nearly 20 years later, the two guys who led the last major shift are back at it again. Suhas, with “soft silicon,” and Gordie, with “reconfigurable silicon,” are working independently to pioneer the “chipless” semiconductor movement, made possible in equal doses because of their visions and hard work and a convergence of process technologies, transistor densities, and cheaper/more powerful software tools. Charter has put $8 million behind their respective companies-Suhas’s Cradle Technologies and Gordie’s Quicksilver Technology-with the belief that this shift is as fundamental and inevitable as the fabless revolution before it. It’s an off-the-shelf approach to getting the ASIC (application specific integrated circuit) that you need.
This type of breakthrough thinking has been tried before, but reconfigurability was never part of the plan. Some of the companies who architected multi-processor solutions such as Chromatis (now out of business) and C-Port (acquired by Motorola), fell short due to the difficulties they had in programmability (they needed Ph.D.s to do it). They were also confronted with rapid tapering of performance (bus contentions, I/O limitations and memory addressibility issues). Transmeta also uses a similar multiprocessor approach, and it solved many of the programability issues by leveraging Linux, but recofigurability was not its design objective.
In contrast, Cradle and Quicksilver have embarked on a radically new paradigm: making reconfigurability the prime driver of the core architecture, which adds tremendous scope and risk to their projects. Only time will tell if this was a bet worth making.
Off the Shelf
Cradle, a fabless semiconductor company founded in 1998, has invented a new single-chip off-the-shelf multiprocessor platform for packet and stream processing applications called the Universal Microsystem (UMS). The UMS is the first software programmable packet and stream processing platform that permits a broad range of telecommunications and multimedia processing applications including virtual private networks, integrated voice and data routers, networked multifunction peripherals and next-generation set top boxes to be defined entirely in software.
Using UMS, you can take a router that needs 20 odd chips and replace them with a single UMS platform. And, because it is implemented in executable software code, you can provide much greater flexibility to adapt to end-user requirements. The real kicker is that other engineers can use the same UMS chip to implement different ASIC functionalities, such as processing images needed in high- quality printers and copiers.
OEM customers can purchase UMS chips the same way PC vendors buy microprocessors and then load software combinations to create the system functions for their products. With this soft silicon approach, many different technologies, such as packet processing and security acceleration, or voice and data, can be combined and economically deployed on programmable and reconfigurable chips. That will significantly drive down the cost of products and reduce the time it takes to get to market.
With an average complex ASIC development cycle of two years, UMS platform development is more like three to six months. For venture capitalists backing companies that require certain ASICs it further helps drive down the costs of doing business, conceivably requiring far less invested capital.
While Cradle is attacking the problem from the networking side, Quicksilver sees opportunities in consumer electronics. If a car radio were built with a Quicksilver chip, it could be used not only as a radio but, with the right programming, a digital radio, a video player, a cell phone, an MP3 player, GPS, you name it. You simply purchase the modules you want and the chip gets automatically re-programmed. This future may not be far away.
“It’s all about designing a chip that is reconfigurable, manufacturing a chip like a generic part rather than a custom ASIC for a specific application,” says Campbell, chairman of Quicksilver and a VC at TechFarm. As a result, customers have absolute control over their intellectual property. That gives them faster time to market. And, if there’s an error in a chip, it can be fixed quickly and cheaply, because it’s just a matter of reconfiguring the software.
From a venture capitalist’s point of view, what’s exciting about chipless semiconductor companies is not only the potential of the companies we are backing-which are helping create the market-but the potential for other companies to be created with far less capital. For example, a Colorado-based company (which shall remain nameless), is using Cradle’s technology and has been able to bootstrap itself and still build the chip it needs for its own next-generation networking boxes. It has been able to demonstrate production-quality silicon in less than six months. Quite an achievement.
The concept of chipless semiconductors is both a cost and complexity issue. From the customer side, for a high-end set top box, you can have a special chip made at LSI, Toshiba or IBM or you can get a box with development software that runs on a PC and write your own algorithms. To test this latter concept, an engineer at a major consumer electronics company downloaded the software tool and was able to demonstrate a full-blown consumer application in less than three months.
Give It a Spin
Because Cradle creates software rather than hardware, you can load the algorithms on a chip and see how they work without creating ASIC mask sets, which run as high as $750,000 per spin for .13-micron technology. It takes an average of two to three spins for a complex chip.
Moreover, because we now face a world with chips with more than 64 million transistors-soon to be crossing the 100 million transistors threshold-we’re getting to the point where what’s being created is beyond the nature of what logic designers can handle. We need a way to move the design from the logic designers to the algorithm designers. “Soft silicon and chipless semiconductors are essentially necessary now because there’s not enough time in the day to change the world each time you need a new chip,” Suhas says.
And if it will help venture capitalists show greater returns on less capital invested in chip-related startups, all the better for both entrepreneurs and their backers. t
Ravi Chiruvolu, a general partner of Charter Venture Capital, is a regular technology columnist for VCJ. If you’d like to send him feedback or ideas, email him at