Get email delivery of the Cadence blog featured here
As part of writing Fabless: the Transformation of the Semiconductor Industry a couple of years ago, I wanted to cover Chips and Technologies, which I believed was the first fabless semiconductor company. One of the founders was Dadao Banatao, now Managing Partner at Tallwood Venture Capital, and so I arranged to talk to him on the phone. The call was originally scheduled for half-an-hour, but he was having such fun reminiscing that we ended up talking for over an hour until his assistant finally managed to pull him away. If you were waiting to pitch Tallwood on your startup that morning, then sorry.
In 1985, Dado Banatao and Gordon Campbell started a company called Chips and Technologies, the first fabless semiconductor company. Campbell, along with Bernie Vonderschmitt of Xilinx, are credited with pioneering the new, fabless, business model. Disruptive new concepts usually meet with resistance by the establishment, and in fact Dadao told me that they had a hard time raising money because VCs couldn't comprehend a fabless semiconductor company. Even his friends told him it "wasn't a real semiconductor company." In fact the first $1M was raised from a real-estate investor! Only after they were further along were they able to raise another $3M from various Japanese investors, including the large Japanese conglomerate, Mitsui Group. They made a technical decision to use gate-array, rather than standard-cell design, because the booming PC market created a sense of urgency in getting their chipsets to market. They chose the industry-leading gate-array technology from Toshiba, but then found that their design was too large for even the biggest gate-array. Their solution was to partition the design into two chips: one logic CMOS gate-array from Toshiba, and a separate bipolar chip with all the IO drivers that Hitachi fabbed. Hitachi had a lots of available fab capacity due to the semiconductor downturn at the time and were somewhat desperate for business. C&T filled up their fab lines completely, and got unbelievably low prices.
C&T’s business took off fast. C&T sold chipsets for IBM’s PC-XT and PC-AT, and made $12 million profit in the first four months after their chipset was introduced. By the time they had their IPO in 1987, just 22 months after opening their doors, they still had $1M of their original $4M investment in the bank. The fact that Mitsui was an investor turned out to be fortuitous because Toshiba and Hitachi were part of the Mitsui keiretsu. This let them order from Toshiba and Hitachi without having to pay up-front with working capital that they didn't have. Mitsui financed $50M in inventory. C&T products dominated for about two years before their competitor, VLSI Technology, started selling its first chipset. The bipolar chip turned out to give C&T an advantage, since at that time electrostatic discharge (ESD) protection on CMOS was in its infancy and was still a potential problem. This let C&T make the claim of better reliability compared to VLSI's all-CMOS solution. Three years later C&T’s chipset was all CMOS too, but by then ESD protection was up to 20KV and those issues had gone away.
Dadao left the company in 1989 to launch another start-up, S3 Graphics, which focused on graphics processors. They again used the gate-array technology to get to market fast. They looked around for the biggest arrays and found what they needed at Seiko-Epson. Dadaobana key invention at S3 was a new interconnect, a local bus, to move data between chips faster. They called it Advanced Chip Interconnect, which later became Intel’s PCI and eventually the industry-standard PCIe.
C&T sold chipsets to Dell very early, and they rode the PC rocket together. Compaq was the top PC maker at the time, but they didn't yet believe in using chipsets. Before long, Taiwan, Korea and Japan were all making PCs and Compaq couldn't compete, so they switched from single chips to chipsets too. Interestingly, at that point in the industry history, C&T were making more on each PC than Intel was.
Gordon Campbell, who had come from Intel and had a previous startup called SEEQ Technology, left the company in 1993 to found 3Dfx Interactive and has been involved with many other successful companies since then. By the time he left C&T, the company had been suffering from lower sales and big losses. The new president and chief executive officer, Jim Stafford, managed to get them back to profitability by restructuring the management and some product lines.
C&T went on to become one of the largest suppliers of graphics processors for notebook PCs. They began to work with Intel and Lockheed Martin on a new graphics chip for desktop PCs and workstations, the Intel740, which was unveiled in early 1997. This collaboration proved to be more than was evident at the time. By July 1997, Intel’s all cash offer of about $400 million to acquire C&T was a done deal. This was Intel’s largest acquisition ever at the time.
The impact of the new fabless business model started by C&T can’t be understated. This is by far the dominant model for the semiconductor industry today. When fabless companies started to be founded was still the era where, as Jerry Sanders famously said, "real men have fabs." In a wonderful bit of irony, Jerry Sanders's AMD would be the first major IDM to spin off its foundry operations (to become GF) and go fabless. I think that in that era, being fabless was seen as a low-capital way to get started, and that once companies became big enough, everyone expected they would get their own fab. In fact, no company seems to have done that. Instead, companies like Qualcomm and Broadcom entered the top 10 semiconductor companies using foundries for manufacture, and nobody is expecting them to get a fab any time soon. In the early days, there were no foundries like today, so fabless companies were buying excess capacity from semiconductor partners. It could be very profitable for the manufacturers—I remember at VLSI that we (embarrassingly) made more money per wafer on foundry than we did on ASIC, where we had to do a lot more work. The FPGA market was entirely fabless, since Xilinx was another company with a reasonable claim to being the first fabless semiconductor company, created in the pre-foundry days. Actel, Lattice and all the rest were fabless too. You can discuss among yourselves whether Intel acquiring Altera means that Altera now has its "own" fab.
Sign up for Sunday Brunch, the weekly Breakfast Bytes email.