• First VP Marketing. Then became Director Systems Engineering & Product Planning
• Designed the Mpact Media Processor microprogrammed Peripheral I/O Bus (PBUS)
• Designed Chromatic’s first PCI board product
• Oversaw development of PC Motherboard reference design
• Designed a software distribution and copy protection approach
• Gave company presentations at CompCon, WinHec, and Computer Elements
• Authored articles for EETimes, OEM Magazine, Microprocessor Report, and IEEE Micro
• Oversaw glue ASIC development for product cost and board footprint reduction
• Provided RAMDAC and SRAM design resource for Chromatic’s silicon partner (LG)
• Many customer and partner visits both domestic and overseas (primarily Japan)

While at SuperMac (where I worked before joining Chromatic) I initiated the design of one of the industry’s first – if not the first – Rambus based graphics cards. Through this development I came to know the Rambus guys (where I was something of a hero I was later told). This relationship took me to Chromatic Research (originally called Xenon Microsystems) when I left Supermac in March 94 as there were strong ties between Rambus and Chromatic, including the fact that Mike Farmwald was both co-founder of Rambus and a founder of Chromatic, which meant the Chromatic media processor chip would use Rambus memory chips.

Chromatic’s mission was to augment the limited multimedia capabilities of the Intel CPUs of the day with the addition of a “media processor” that would either be on the motherboard with the CPU, or located in a separate PCI plug-in card. This media processor, called Mpact (originally called the VSP), was marketed as being capable of the following seven multimedia functions :

1. MPEG Video (which quickly evolved into software DVD playback)
2. 2D Graphics
3. 3D Graphics
4. Audio
5. Fax/Modem (V.34)
6. Telephony
7. Videophone

I joined when the team numbered about 20 and Chromatic was located at an incubator complex at 2083 Landings Dr in Mountain View, now just a stone’s throw from the GooglePlex. My initial role was as VP Marketing, and during my relatively short tenure in that role I had a meeting with Kevin Harvey of Benchmark where he asked me if I wished to formally take the VP Marketing role as they thought I could handle the job, but I told him I just didn’t think marketing was my calling and so after filling the role for six months or so the company hired Siva Kumar as VP Marketing. That may have been a mistake on my part as I created a pretty hard ceiling for myself to break through at Chromatic. One of my enduring contributions though as VP Marketing was the logo of the hand touching the water (which can be seen on the left of this page) creating the symmetric propagating wave pattern. My goal there was to create an evocative image that “humanized” the technology and connected the human audio/visual experience with the underlying technology enabling it. The exclamation point Mpact logo was created later under Siva’s tenure.

I shifted to Director of Systems Engineering and Product Planning. In that role I defined the I/O architecture for Mpact’s Peripheral Bus Interface to video, modem, audio, and RAMDAC etc chips, designed the PCI reference board and oversaw the development of a motherboard reference design in partnership with a third party (Anigma, based in southern California). I was actively involved with a number of potential OEM customers, including Gateway and Dell.

Since Mpact had to connect to a wide variety of unknown peripheral chips (display RAMDACs, audio codecs, modem chips, video encoder, video decoders, UARTS, game hardware interfaces, etc), the interface had to be extremely flexible. The only approach that made sense to me was a microprogrammed “bit banging” interface, where all the signals could be wiggled with fairly arbitrary timing using small microprograms. I specified this architecture and it ended up working quite well.

The design of Chromatic’s first product (as well as system software development and evaluation board) was a technically very challenging 2/3 length PCI board. This board incorporated fast noisy high speed digital signals (display/RAMDAC interface, PCI bus, 700Mhz controlled impedance RAMBUS channel, the Mpact chip itself, etc ) all in extremely close proximity to functionality that required a very “quiet” environment necessary to achieve high Signal-to-Noise (SNR). Examples there were the V.34 telephone modem interface and codec, the audio codec, and the video signal input path. I designed this board from scratch but used a third party to route the design after I carefully specified component placement and all the the layers and ground/power plane treatments. I probably learned more designing this board than any of the previous chip designs I had done. A labeled photo of the Mpact board is shown below on this page.

I was even then a big proponent of instilling a power conscious design ethic in the company. My interest in low power design started with the battery driven clock chip at Apple, but subsequently matured during my time as principal hardware architect of the Apple Newton – where low power consumption was a top design priority. I had assembled a survey document of low power techniques at SuperMac, but it was not nearly as apropos there as it was to the Mpact processor designs. I evangelized the doc within the company, and include it below for the interested reader. Even though it is ~20+ years old, not much as changed as it is pretty much basic physics. What has changed (improved) are attitudes, methodologies, and tools.

Chromatic’s first fab partner was Toshiba, and Wes Patterson, Chromatic’s CEO and former eVP and COO of Xilinx, was a very strong believer in partnerships. Too strong perhaps. The choice of Toshiba was no surprise given the pre-existing strong relationship between Rambus and Toshiba.

The Mpact architecture was the brainchild of Steve Purcell, who was a Chromatic founder and CTO and had most recently been a founder and chief architect at C-Cube Microsystems. Mpact was a VLIW (Very Long Instruction Word) architecture, a type of SIMD (Single Instruction Multiple Data) architecture, that was designed to do a lot of things at once using a lower clock rate than was typical for RISC processors of the day (which ran counter to the dictum drilled into us at Berkeley – originally from Seymour Cray – which was “nothing beats a fast clock“). Mpact was likely an evolution of the types of video focused architectures Steve had been deploying a C-Cube. The architecture centered around a strongly multiported memory (4 write ports and 4 read ports) that fed a large number of independent functional units over an internal 800 bit bus. Having been raised up through through the RISC (Reduced Instruction Set Computer) world under David Patterson at Berkeley, where it was inculcated into us that nothing beats architectures that use dense, fast, simple single ported memories – I was wary of a multi-port memories (especially anything beyond a dual port memory). Multi-port memories also had a reputation of being buggy. Sure enough, although Chromatic designed the chip, Toshiba insisted on designing the memory and the first iteration of the chip had SRAM memory problems (pattern sensitivities, etc).

But I think by far the bigger issue due to the architecture was lack (at least initially) of a high quality compiler and the resulting impact (no pun intended) on the ability to scale the software effort. Because the architecture was unusual, perhaps even arcane (the architecture is presented in more detail below in the downloadable article) – given the compiler technology of that time a quality optimizing C compiler was a long ways off. This of course meant that to achieve the necessary performance much of the code, and certainly all the critical inner loops, had to be coded by hand in assembly language. Not only coded by hand, but coded by hand by what I called the “tiger team” – a small group of recently minted PhDs who could coax out that last 15-20% of performance that made the product viable. That team included Avery Wang (who went on to found Shazaam), Andy Hung, Diego Garrido, and others (including Dave Miller, Chad Fogg, Dave de Heer, Brooks Read, Denis Gulsen, and Dan Culbert). In the face of both new and constantly changing multimedia and graphics APIs from Microsoft, along with ever increasing requirements of OEMs, software demands began to eat the company. The breadth of software required I’m sure dwarfed anything experienced at C-Cube.

Chromatic’s business model was unique, was I believe invented by Mike Farmwald, and was known as the “chipless semiconductor company” model. Since pure play foundries such as TSMC did not exist at that time, one had to partner to access a fab (this was back in the “real me own fabs” days), and typically the available partners did not make bleeding edge fab facilities available to 3d parties. This model also resulted in more margin stacking than is seen today with the pure play foundries. Hence the thinking was to develop the chip and ship tooling that is ready for fab to your partner and from that point on they own the chip from a marketing, pricing, and sales standpoint.

So how did Chromatic make money you ask? From the sale of the software than runs on the media processor. This can be seen in the diagram below that is a excerpt from the original business plan:

So what could go wrong? Quite a bit as it turned out.

1. Software development burn rate eats the company: The burn rate at the company to support all the software development as well as support a high end chip design team developing designs for different processes at multiple partners could turn out to be much higher than expected. This in fact happened as when I left the company in early ’97 the headcount was ~260!, the majority of which were writing code.
2. Market killing pricing by your fab partner: As the chip pricing is our of your control, your fab partner could price the processor much higher then you expected or may not market the chip as aggressively or in the way you believe is best, creating a significant barrier to adoption and putting strong price pressure on your software. That in fact is what happened when Toshiba initially priced the processor at $90 (when the business plan called for a price in the $30-$35 range). It wasn’t until LG was brought on as a second source over a year later that processor pricing reasonably aligned with the business plan.
3. Value slicing and dicing by your customers (the OEMs). To command premium pricing, your customers really needed to buy into the value of a “full package” of functionality. But if all they really want is one or two functions, such as your software DVD playback, and for that they will only pay a small fraction of the full functionality ASP – then you are in trouble. And this of course is what happened.
4. High table stakes to sell into the OEMs: You could underestimate the amount of work you had to do to get OEMs to adopt that part. In fact not only did you have to design the media processor that you gave to your fab partner, but you had to provide turnkey reference designs for the OEMs, including both PCIe board designs as well as motherboard designs.
5. Theft. I was quite concerned that Chromatic would be unable to track software shipments, particularly when board production was moved offshore (Taiwan at that time). Since the software was more like firmware in that it shipped with the board, and there wasn’t a unique activation code shipped with each copy of Chromatic’s software, there was very little barrier to suppliers simply duplicating the software and shipping product without paying Chromatic. I was told after I had left Chromatic, and privately, that this in fact did prove to be a problem. For this reason I had added a single pin interface to the peripherals bus on the Mpact to support an inexpensive 6-pin Dallas Semiconductor unique ID chip. With a unique ID associated with each Mpact, it was possible to place a Chromatic controlled PC on the supplier’s assembly line and have that PC “phone home” to Chromatic to enable the authenticated download and activation of software on each board. While the capability was there it was never fully developed or used, as no one understood it after I left and it was a barrier to adoption. Not adopting it however apparently proved somewhat of a barrier to revenue. The reader should remember that this was in a time before everyone was routinely connected to the internet – so we couldn’t have every shipped system simply phone home to Chromatic’s server to authenticate itself.
6. Requirements evolution/escalation: While Mpact was really designed as a quite respectable 2D graphics engine, and could do 3D, it was not really a high performance 3D engine. High performance 3D is a different animal. An animal for which the Mpact-1 architecture was not well suited. For one thing you need at least a minimalist floating point representation, supported by the hardware, to prevent rendering artifacts. Companies like nVidia and ATI were developing very specialized purpose built streaming floating point 3D graphics rendering engines by that point in time.
7. Moore’s Law: The inexorable march of Moore’s law meant that just about all functionality, with the exception of 3D graphics and some high speed networking or communications baseband signal processing functions, could eventually be handled by the Intel CPU. This latter issue was not lost on me in the selection of my next project, which was a baseband communications processor. These last two issues drove Chromatic to conceive, in the fall of ’96, of a project called Tapestry, that would integrate an x86 binary code compatibility and much higher performance 3D graphics into a product called “A PC on a Chip”. Tapestry proved a bridge too far.

The original business plan (available for download below) called for an average ASP of $7/copy for the full functionality Mpact software suite, but that assumed a company burn rate that turned out to be far smaller than reality. A year or so after the original business plan plan, the ASP per copy rose significantly and the number that I recall was a price in the $15-$20/copy range was the socialized price within Chromatic, which was a level that unfortunately the market would not support. Being probably one of the few employees that actually read the original business plan, I was becoming very concerned about the “slicing and dicing” of the value proposition being done by the OEMs and what that was doing to the ASP we could command.

Why did I leave? Reviewing the notes I made prior to my exit meeting with Wes Patterson, there were a number of reasons, but the two cited below were dominant:

1. I had been a “good trooper” owning the system engineering and system prototyping/board design tasks, giving presentations, writing articles, writing MRDs, going on sales calls to Demo, etc. – and although designing the Mpact multimedia board was non-trival, and I learned a lot, I felt I was becoming typecast and was running into an advancement ceiling. I told Wes “without have a structural position commensurate with one’s contributions, there is no appropriate long-term recognition”.
2. I was also also convinced the business model would fail unless we “took back” the Mpact chip business and recovered the fabless semiconductor component of the business, which I knew would never happen under Wes, and probably legally we couldn’t unwind anyway. I clearly remember walking into Mike Ofstedahl’s office (Mike was VP Sales), and he was pumped, and when I asked why he said, “we just closed Packard Bell” – and I said “great – what is the ASP?” and he said “$4”., at which point I thought to myself “game over”.

The time I spent with Mike Farmwald while at Chromatic, along with the exposure to Benchmark Capital (as an investor) let me to my next engagement, which was as an EIR (Entrepreneur-in-Residence) at Benchmark Capital: BENCHMARK-CAPITAL

TAKEAWAYS:

1. If a VC offers you a VP slot, think twice about saying no even if you don’t think “it is your calling”. You will learn a lot regardless, and if you hanker for greater responsibility saying no will set you back a few years.
2. If you come to believe that “this dog don’t hunt” from a product or business perspective, then think very hard about whether you should stay at the company and whether you can keep you heart in it. When my wife asked me “why do you keep doing these startups?” – I told tell her “because its where I get to believe”.

MEMORIES:

I recall multiple trips to Japan with Wes and other senior staff. Potential customer and partner meetings, Fab tour(s) (probably Toshiba). We were hosted on at least one occasion by Takahiro (Tom) Kamo, President of Rambus KK and Intel Japan who was also on retainer with Chromatic. He was a wonderful person. I recall dinner and red wine with Tom’s wife Cindy at their house, and playing golf with Tom at Toyosato Golf Club where I believe he was a member. Tom loved golf – although I found the Japanese style of golf play, with lots of girls in white hats and white gloves wanting to help you and show you where your ball was, etc, a bit disconcerting.
I vividly remember departing a meeting on the 3rd floor of a 3-story building at Toshiba, and the group decided to take the stairs down. We ran into one of the older women who brought tea up to the meetings. She took one look at us and exclaimed “ohh Kamo-san” looked down and tried to shrink to invisibility in the corner of the stairwell. Coming from the US and raised in the midwest, that level of deference to, and honor of authority was an eye-opener.

I made several trips to Gateway when they were still in South Sioux City South Dakota. I recall one time scrambling around at the office trying to find a jacket for Mike Farmwald as he was about to get on a plane to Gateway and he seemed to have no clue about what January meant in South Dakota. Flying into Eppley field in Omaha was a recurring deja vu moment for me as I grew up there.

I recall flying down to LA on very short notice to rescue the very first Mpact chips from Japan that were stuck in customs.

I made multiple out of state customer visits with Wes, where I would bring a demo system and run the demo – including Dallas, and then Houston where we met with Michael Dell whose office at that time was in a nondescript one story industrial tilt-up building north of Houston. He was quite congenial, and seemed very young.

Chromatic Slide Show:

1. Company Team Photo, Mpact First Ship Oct ’95
2. Mpact-II DVD Board
3. Mpact-II Graphics Only Board
4. Gateway Mpact DVD Board