Specialized Chipmakers, the Modern Architecture Play

As soon as I set foot in the Robbie Stephens Semiconductor Conference, I got the willies. It reminded me of the 10 years I spent as a sell-side analyst tracking semiconductors, a.k.a. "riding the roller coaster." These chip stocks were awesome short-term trading vehicles but had the unfortunate habit of going down when they were supposed to go up -- and vice-versa.

For much of that time I was pitching Intel (INTC) as a long-term investment, but I would spend all my time figuring out whether it would hit a brick wall the next quarter. Over the years, Intel was far more than a 10-bagger, but the game was all about figuring out the cycles, and the ride caused severe whiplash. I came to hate semiconductors, and never wanted to think about them again. I figured I had better start doing something else so I wouldn't end up with gray hair and ulcers.

I was reminded of all this last week, when Micron's (MU) stock shot up nearly 18% after investor-relations exec Kip Bedard mentioned that spot prices were above $6 and had been there for a few days. That's right, days. In the past 12 months, the stock has been as high as 80 and as low as 20 and now is near 70 again: living proof that you rent these stocks -- you don't own them.

In our fund's quarterly newsletter, we like to report the makeup of our fund by segment: software, network equipment, etc. Lo and behold, 40% of our portfolio is invested in companies considered semiconductor companies. Another 20% is invested in companies whose secret sauce is a semiconductor with software running on top of it. Thus I'm back to being up to my eyeballs in chips: Am I nuts?

Maybe, but I don't really consider any of these outfits semiconductor companies: I consider them intellectual property companies that happen to deliver their value-added on silicon. Their value is in their architecture, not the commodity silicon it resides on. If my partner and I can find next-generation chip architectures that network equipment vendors and telecom vendors need to implement their systems, we can invest without all that much thought to where we are in the semiconductor cycle, and not give much thought to where commodity chip prices are headed.

A case in point is our investment in Hi/fn (HIFN), which was spun out last December from Stac, now Stac Software (STAC). Stac, if you remember, sold compression software for PCs, but was attacked by Microsoft (MSFT) and eventually won more than $100 million from Microsoft for patent infringement. Less well known is that Stac had a semiconductor business. Quantum (QNTM) approached Stac and said it needed a lossless compression scheme for DLT tape backup drives, to double both their capacity and their speed. Stac took its LZS compression scheme, which ran too slowly as a piece of software, and implemented it in silicon.

Since then, using the same logic, Stac figured out that it could also offer hardware-based encryption. A hot area in the network space today is virtual private networks, or VPNs. A Pentium III can handle encryption for a 56K modem line and maybe even a 384K digital subscriber line, or DSL. But businesses that want to implement virtual private networks for their higher-speed lines like T-1 or T-3, and soon, optical-based connections, need high-speed hardware encryption. Hi/fn not only has it, but ties compression and encryption together to optimize encryption specifically for VPNs. So we owned Stac (no longer) and now we own Hi/fn. It has been a great stock this year as network equipment companies such as Ascend (ASND) have begun designing Hi/fn's chips into their products and ordering them in large numbers.

So do we own a semiconductor company? Sort of. Do we own an architecture company? You bet. Are there cycles to worry about? Yes, but less so than for Micron and that wacky DRAM market.

The combination of software delivered on silicon is also very powerful. One such private company we are talking to considered giving away its silicon architecture and then just selling its software, to the likes of Cisco (CSCO) and Lucent (LU), etc. That company quickly changed its mind when management figured out that Cisco wasn't really interested in licensing software for very much money -- say $1 or $2 per unit -- but would be willing to pay $50 for a chip. There is something about the deliverability of semiconductors that enhances their overall value vs. just software.

There are tons of examples of architecture companies that sell silicon. Broadcom (BRCM) and PMC-Sierra (PMCS) are two great large-cap examples. You have to dig a bit deeper into the small cap space to find more such cases. I've talked in the past about our investment in MMC Networks (MMCN), which sells network processors to Cisco, IBM (IBM) and others. Be forewarned -- a lot of these stocks have had serious runups, so be careful.

Over time, the value of these architectures delivered in silicon will wane. There used to be a great market for VGA graphics controllers, until it got too crowded and squeezed the value to almost nothing, making intellectual property-deliverers into commodity vendors. This could also happen in the networking space -- for those that don't continue to innovate.

>To order reprints of this article, click here: Reprints