Of Chips and Dips

I have to admit that I am sucker for semiconductor startups. It’s a bit of an addiction that started when I helped lead development of a ASIC-based Ethernet switch skunkworks project at Chipcom a million years ago (1993 to be more precise), and has continued into my career as a venture capitalist.

The common denominators of successful semiconductor companies are what attract me: big markets, the need for deep domain-experienced teams with idea-innovation, execution and manufacturing skills. With all due respect to my partners who invest in Web 2.0, the simple act of producing a semiconductor that works as specified is just plain hard, can take several years and can cost many millions of dollars before we even know if it works, let alone whether customers will buy it.

We’ve been rewarded handsomely in the past for taking the risks of funding such companies, with IPOs and M&A transactions providing great returns from the late 90’s into early 2000. Chip geometries were 250nm and 180nm and costs of getting to market were in the $5-10m range, the datacom and telecom markets were thriving and therefore IPO and M&A markets were energized.

That was then. The dollars needed to fund production of a semiconductor have escalated exponentially over the past few years, driven by the inverse relationship between nanometer-scale chip geometries and total cost of development (software tools + fabrication set up charges + materials). Estimates of developing a complex 90nm ASIC are now between $20-$30 million and while this decreases in real time as the experience curve evolves, the industry keeps pushing to even smaller line widths – with 65nm and 45nm starts already happening. It should come as no surprise that most VC-backed startups have gone after more challenging projects – highly-complex integrated digital and mixed-signal products – e.g. network processors, Ethernet switches, security processors, multi-core high performance CPUs. (Why take on the mundane?). But with relatively few exceptions (Advanced Analogic, Ikanos, SigmaTel) the IPO window has been shut for a good long time for component companies. And while there have been a handful of M&A transactions over the same time period (Oasis, Passave, Quake, Sandburst, Sandvideo) scant few have crossed the $100m mark in value, while the total amount of capital raised has been in excess of $50m. (and I won’t depress you with the RIP list of dead semi startups)

This is where the calculus breaks and the status quo for semiconductor startups needs to be reconsidered. If it costs at least $25m to get a first product to market, and an additional $25m for a second product and dealing with any unforeseen issues … then $50m doesn’t seem that outrageous. If the outcome for such companies under reasonable assumptions will be in the $100m range, then quite frankly this is not a good venture capital deal and guys like me shouldn’t be funding such companies.

The Semiconductor Development Cycle: a slide from a presentation I made to my partners in March 2006.

So if my math is right, why are some venture capitalists still investing in highly-complex digital semiconductor startups?

The worldwide revenue for semiconductors is in excess of $200b annually across all segments. Big market. Some venture capitalists invest in highly capital intensive projects because they see the promise of the next Broadcom or Marvell, and usually costs money to win big. Personally, I am not ruling out ever investing in complex digital semiconductor companies again, however my current perspective makes me look more longingly past the first round (maybe even the second round) at companies that have made it past development risk and have established meaningful contact with important customers. (*Note: as the saying goes, management reserves the right to change the menu without prior notice; I am always testing my market assumptions).

If I am really this negative on the market, why am I still investing in semiconductor startups at all?

Despite the default focus within the venture ecosystem, the majority of products comprising the $225b annual market are not complex high-speed digital at all, i.e. the calculus I present above is not valid in these other cases:

• Some of these chips are not highly complex or deeply integrated with many functions. (simple chips are cheaper to build)
• Some of these parts don’t require tiny and expensive geometries. (bigger chips can be fabricated faster ad easier),
• Others aren’t 100% digital, but instead are mixed-signal (analog + some digital) or all analog (analog is cheaper than digital).
• Markets are broader than traditional VC-startup foci, i.e. not telecom, datacom, wireless. (more customers to sell to)

How are others reacting?

According to the PWC Moneytree Report, semiconductor startup financings peaked in Q3 of 2000 at $1.1 b and bottomed out in Q4 2002 at $231 m. Great – others recognize the issues in this segment? Nope; despite the obvious conclusions I present here, VC financings have steadily risen to $554 million in the second quarter of 2006.

Conclusions

So my conclusion is (1) there are many markets that need innovation at the chip level, (2) ad not all these opportunities require massive investment to get to proof points and (3) these are the kind of companies I’m looking to fund in first round financings. We’ll have to just have to wait see if I am right …