You like to poke at Cube Management, this says blame the board?
Interesting reading about Prakesh.
High ambition: Ambit Design Systems, a small start-up, reinvents chip design technology
Sukumar Ramanathan
05/31/1998
India Currents
Page 26
Copyright 1998 SoftLine Information, Inc.
high ambition: Ambit Design Systems, a small start-up, reinvents chip design technology
Prakash Bhalerao is quite possibly the most articulate executive you will ever meet. And definitely the most enthusiastic. As you sit opposite him in his office at Ambit Design Systems, you are mesmerized by the flow of words, charmed by the anecdotes, taken by his energy. You find yourself seized with wonder at things you didn't even know existed an hour ago, arcane processes, mystifying products, strange acronyms. You are enthralled that this man has devoted his life to carving minuscule lines on a thumbnail of silicon, lines so narrow that a thousand of them would fit inside the thickness of a human hair. And he is so good at exposition that he is able to make you understand why infinitesimal lines are necessary for your six-year-old to guide Super Mario to the princess.
Ambit is in the business of electronic design automation. This is a fancy term for software that figures out how to lay the circuitry on a chip so that it does what you want it to do - whether guide a jumbo safely down into LaGuardia or keep the rice warm for your spouse working late again at the office. This software has been through three incarnations.
First, ranging from approximately 1985 to 1990, it was used for lines to be etched onto the chips at the micron level. The companies that ruled this marketplace were named Mentor and Daisy and Valid. Then, the Intel 386 came out, chips got more complex, and starting around 1991, the lines needed to be etched in submicrons. A new generation of companies sprang up to win in this marketplace, the most prominent being Cadence and Gateway and Synopsys.
Today, the Pentium chip has over five million transistors, and the lines have to come in at a third of their previous thickness, a previously unimaginable 0.18 microns; Prakash Bhalerao is confident that his company's software is going to be the market leader in this third era, that of deep submicron technology. He certainly has a head start.
EDA software is horrendously complex, and even as a startup, Ambit's flagship product, BuildGates 2.0, runs about two million lines of tightly written code. It is important to understand why the code is so detailed.
Some years ago, chips got so complicated that no one person could keep the complete design in their heads and figure out what each input signal would result in. The entire design had to be broken up into blocks, and an engineer assigned to each block. You designed your little piece of the chip as perfectly as you could, using everything you had learned about logic design and transistor circuits in graduate school. On the huge schedule stapled to the wall, this stage was called "Schematic Capture." After you laid out all the little gates, you took the help of the expensive software package on your Sun workstation to calculate how best to place the gates so that it took as little physical space as possible. On that giant schedule, this stage was labeled "Place and Route." You prayed daily that everybody else in your building was as painstaking as you were, so that at the end, when you put all the little blocks together, and reached the stage called "Verification," the entire chip would work perfectly.
Of course, this never happened, and then there would be tense meetings and tight lips and coffee that tasted like mud at two in the morning. The days would meld into nights, and the weeks into months, and each sample chip that came back from fabrication test would fail in some new way. You would worry constantly, praying that this was the fix that would solve everything, that would flash all green and ship finally and stop the CEO from sending status requests and buttress the stock price and save the company. And then just as that happened, and you wanted to take a month off, and take the wife and kids to Ooty, six weeks thinking about nothing more worrisome than when to have the next cup of tea, your boss would come into your office, and tell you to cancel all vacation plans because you had just been assigned to the new chip called Cheetah.
This is why running a good Electronic Design Automation package is very important.
Bhalerao has been in the chip business for almost as long as there has been a chip business. He started off at Digital Equipment Corporation in 1970, fresh from a master's degree at Worcester Polytechnic. He did not know it then, but DEC was probably the best place in the world at the time to learn about chips. It had succeeded magnificently in defining a new segment called minicomputers, and its engineers were kept busy cranking out one bestseller after another - the PDP-8, the PDP-11, the VAX 11/750, and the VAX 11/780. The group was headed by the legendary Gordon Bell, one of the best computer designers that ever lived.
He started out like all the people in the cubicles next to him, laying out Boolean equations and truth tables and adder circuits. He rose in the manufacturing organization, helping put in place a smart mechanism for keeping costs down. The engineers coming out with new designs could ask all the manufacturing lines within DEC to bid for their business. If the winning line bid $300 per chip but was able to wring efficiencies from its processes and actually make the chip for $200, then it was allowed keep the extra $100 and invest it in itself so that it could be even more competitive for the next bid. Many years later, management guru Tom Peters would dub this intrapreneurship. By the late 1980s, Bhalerao was managing DEC's entire chip business, in charge of 2,000 people bringing in $1.5 billion in revenues.
There was something nagging him, however. He wondered how much of his success was attributable to the fact that his company bestrode the computing world like a Goliath. The real challenge was to be in a smaller place, one that would be an acid test of his entrepreneurial abilities. So he decided to look for the right opening.
The chance came from an unexpected place. In his capacity as vice-president, he had initiated business with a small company called LSI Logic. He like the opportunism showed by its founders Wilf Corrigan, Bill O'Meara, and Rob Walker. They had once approached him for business, showing their plans for a chip based on Emitter Coupled Logic. Hah, he had told them. ECL was old hat. He needed chips based on this new process called CMOS. Wilf and Bill and Rob looked not the least bit nonplused. They asked for a few days, saying that they had to fiddle with their business plan. A week later, they were back to talk about the new LSI, one that had been reincarnated. Presto chango, it was now a CMOS company. They got the business, Stage One in an eventual pat to $1 billion in revenue.
Now, in 1990, Bill O'Meara had joined a small company called C-Cube. The company had been founded by Ed Sun, a man with a reputation as a serial entrepreneur. It had originally been in the business of data compression on hard disks. But hard disks were growing in capacity at such astonishing rates that even the best algorithms could not compress the growing data without losing vital pieces. They looked around for some medium that could retain useful information even if some bits were lost. They settled upon video. The human eye is so good at extrapolation of images that it can make up for the loss of a few pixels here and there in a video display. So video compression it was, modifying the algorithms such that you could tune your compression rate. The less you compressed, the less your picture degraded.
C-Cube was small, 20 people, almost all of them engineers. Bill flew out to Maynard and threw down the gauntlet. Prakash had always said how he'd like to run something small. It didn't get any smaller than this. How would he like to run both Engineering and Product Marketing? There would be no limousines, no offsite meetings at country clubs, no helicopters ferrying him to the airport. But there would be lots of stock options and the chance to actually do something he had talked about for years. It was a seductive offer. Prakash accepted.
C-Cube would be in for a heady ride. Sequoia, one of the best-regarded venture capital firms in Silicon Valley, warmed up to the notion of inexpensive chips for decoding video signals on the fly, and backed the company. One of its partners, Mike Moritz, talked up the company in Newsweek magazine. Huge markets like PCs and satellite receivers danced in front of their eyes. The chips were devilishly hard to design, especially the piece that encoded the video. But Prakash ran a tight ship, and the company was among the very first to market. He topped this feat by brokering a deal with JVC Electronics, one that would eventually result in sales of four million units a year.
At the moment of success, when everybody from Compression Labs to Thomson CSF was clamoring for their chips, Prakash started to have misgivings. They made their chips to sell for about $1,500. Six of these chips running in tandem with their software and a few supporting chipsets was enough to make a videoserver, a device that decompressed signals and delivered video on demand. Other companies were buying these components from C-Cube , mounting them inside a chassis, slapping their own name tags, and selling the servers for a quarter of a million dollars. Why piddle around with the small margins of a chip business when the 85% gross margins of a systems business was starting them right in the face?
Here though, he ran into a buzzsaw. You can be either a chip company or a systems company, not both. It takes a different kind of thinking, and a different kind of sales force. Not even Intel tries to sell its own PCs. The venture capitalists were forceful about this. But he dug in his heels. Eventually, Ed Sun and he both left. His options package was quite generous, removing money from being a major issue from his subsequent ventures.
Ed Sun went off to China and set in motion a chain of events that has deeply influenced the way that a quarter of the world's population seeks entertainment. The reason that video cassette drives are hugely popular in China is because of what he did there in the early nineties. He is further responsible for frightening displays of karaoke singing in the Far East. And finally, his work has unwittingly resulted in millions of cheap and illegal copies of Titanic being available in Shanghai a full three months before its theatrical release in the United States. Ironically, he helped C-Cube much more after leaving it than when he ran the show.
China in the early nineties was coming out of decades of cultural repression. Deng Xiao Ping was slowly embracing capitalism and had even said in a speech that it was glorious to be rich. Shanghai's Nanjing Road was beginning to rival the Champs Elysees for opulence. And a third of a billion peasants were moving out of grinding poverty into a lifestyle of refrigerators and gas stoves and televisions.
But Chinese TV programming was drab. There were only so many nights you could listen to a bland announced read out the rice production figures for Hunan. The new middle class wanted ninjas and warlords, damsels and song. And Ed Sun knew just the device that would fill this need. Enter the VCD, a cheap box with a C-Cube chipset that played both video and CDs. Enter perfectly copied Hollywood movies made in a hundred fly-by-night shops and sold for a dollar each. Enter laser karaoke and singalong parties. Five hundred manufacturers sprang up overnight in China. And they had but one source for their chipsets. It was a sales bonanza that caused the executives at the company to marvel at their good fortune.
Prakash Bhalerao was by now off to his next act. In his years at DEC and C-Cube, he had gotten a gut feel that the world's insatiable need for larger and more complex chips was going to open an enormous opportunity. The first generation of chip architecture, with four inch wafers, was during a period when designers actually drew their circuit gates with pen on papers. If you took the case of a high-volume chip that retailed for $5, and if you assumed that an engineer working at full speed could do 50 gates a week, the complexity of the chip that a small team could design in a year was of the order of about 10,000 gates.
When wafers got to be six inches in diameter, the design software companies began to offer reusable libraries of the most common gate designs so that you could plug them together like Lego blocks. This helped engineers be about 10 times as productive, and allowed chip complexities of about 100,000 gates.
At eight inches, which was going to be the standard starting in around 1998, Prakash foresaw that microprocessors would use about 10 million gates. The need for computing power was endless. With a Pentium II, the average user now had more processing ability on his desk than was used for the entire Apollo 11 space program. yet, he was still forking out money for faster machines. To match this need for speed, engineers would have to be a hundred times as productive. People were in serious discussion about the dramatic increase not just in the rate of change but in the rate of rate of change. Chip designers needed to have gate libraries built to higher and higher levels of abstraction. He became the president of Silicon Architects, a company that filled just this need. It was bought by Synopsys in 1995.
As Senior Vice-President within Synopsys, he played the part of gadfly. He told anyone that would listen that the new world needed a complete new design automation architecture, one where all the old assumptions would have to be thrown out and new code written from scratch. When his call feel on silent ears, he began to scout around for an enterprise suitable to realize his vision.
He found it in a startup called Ambit Design Systems. It had top-flight software architects like Devadas Varma from Mentor and Krishna Belahale from IBM. It had received angel funding from Suhas Patil and Kanwal Rekhi, stalwarts of The Indus Entrepreneur organization. It had even raised $5 million in first round funding from Information Technology Ventures. Prakash joined in February of 1997 as the CEO. In short order, he raised another $19 million. He oversaw two releases of the product and the organization's growth to nearly 100 employees.
Silicon Valley is the perfect tested for hot technology. When BuildGates came out, chip designers were abuzz about the remarkable speed of the software and its ability to handle very complex designs. Synopsys dominates the segment, with over 90% of the market. A memory controller that took 12 hours to run a synthesis on Synopsys's code took less than a third the time with Ambit's. Sensing that this upstart was causing a stir, Synopsys was forced to play catch-up by releasing three consecutive versions of its flagship product in a single year. It will lost in a number of competitive benchmarks.
Sun Microsystems, Lucent Technologies, and IBM have all signed on as Ambit customers. Cisco just designed its latest gigabit chip using the software. Sumitomo and Cadence both took equity investments in the corporation. While Prakash won't disclose exact revenues, he will say it is already in the double-digit millions. His plans call for a public stock offering next year.
In a recent on-line piece, the columnist Robert Cringely talked with complete seriousness about how a Sony Playstation retailing for a few hundred dollars next year will have as much graphics rendering capability as a $20,000 SGI workstation. We increasingly expect realistically-shaded three-dimensional characters even in our video games. The engines that hum in the background for sights like these are chips of bewildering complexity. If Prakash Bhalerao realizes his goals, each of those tiny dynamos will have infinitesimal filaments etched on its surface with software made by Ambit Design. |
