Industry Squibs

A couple of bulletins have been published last week concerning the semiconductor industry, and unlike the recent startup and venture capital news, these are actually on the positive side. The Semiconductor Industry Association (SIA) reported that chip sales for the first have of this year are up a healthy 5.4 percent over the same period last year. The SIA attributes the sales growth mostly to strong sales of portable computers and mobile phones, both of which experienced double-digit unit growth. With emerging markets having significant growth in both of these areas, it is not surprising that the Asian-Pacific region had the highest year-over-year growth at close to thirteen percent, with Europe coming in second at around five percent, and finally the Americas and Japan coming in at close to 3 percent.

While the SIA findings are inspiring, any growth is good growth, the IC Insight findings for the same period are even more interesting. There has been quite a bit of movement in the Top 20 list of the semiconductor sales leaders: The biggest position gainers were Qualcomm, NEC, Panasonic, and Broadcom. The biggest looser hands-down was Qimonda. With a 47 percent sales decline year-over-year for the first half of the year, Qimonda dropped a staggering 12 positions and consequently out of the top 20. The list becomes even more interesting when sorted by growth rate. Of the top six spots, three are occupied by fabless companies, and the top spot is occupied by a foundry where a lot of the fabless guys build their chips, namely TSMC. All of the companies ranked in the top six had impressive growth rates of over 20%. It should be noted that several of these companies are exposed to the fast growing mobile market; however, this by itself is not nearly enough, as TI who is usually also a strong player in this market did not experience any growth over the same period - very surprising. There has also been a lot of talk regarding AMD and the asset-light strategy that the company intends to pursue. Freescale has adopted a similar strategy over the last few years, however the company’s 3 percent growth rate shows that going asset-light by itself is not enough, or at least that it might take some time for the results to show. One thing is for sure thought, with the fabless companies doing so well, and with even more companies considering going fabless or asset-light, the foundries must be licking their chops.

We reported earlier this year that 2008 was anything but a stellar start in terms of venture capital for semiconductor startups. It does not take a genius to see that the semiconductor market has been on shaky ground as of late, excluding a couple companies here and there. Nevertheless, running across two articles that highlight the negative and are published on the same day, while addressing the opposite sides of the semiconductor spectrum, is rather depressing.

The first article, titled Why Chip Stocks are Down and written by Steve Tobak, focuses on established semiconductor companies and examines why they have significantly underperformed the market over the last few years even though chip sales have experienced double-digit growth over the same period. According to Steve, some of the lackluster performance can be explained by the recent memory chip glut. Additionally, the dot-com bust which inflated the stock prices of more than one semiconductor company still rears its ugly head to some degree. And while an in-depth analysis of the entire sector is sort of lacking in the article, one observation regarding companies that bucked the trend is quite enlightening: “Proprietary products in hot markets resist negative sector trends, while commodities suffer the most.” This observation is exemplified by companies such as Qualcomm and Marvel who have done rather well for themselves. Interestingly, nVidia did quite well over the same time period too, while Intel stayed about flat - One would think that the fortunes of these two companies would be in lock-step, but this turns out not to be the case at least as far as stock valuation is concerned.

The second article is a commentary by Chris Fisher titled What Price Entrepreneurship? Essentially, the article questions whether starting or joining a semiconductor startup makes sense from an individual’s financial point of view. Chris estimates that a semiconductor startup needs to raise in-between $60 and $120 million these days. Money alone of course does not guarantee that the company will be successful – Montalvo comes to mind as a recent example. Now, if getting this much money was not difficult enough, Chris points out that public markets recently have valued semiconductor companies on the low side at about three times revenues. Add to this the dismal performance of semiconductor IPOs as of late, a lack of interest by established companies in acquiring new ones, and one has to wonder if time spent on a semiconductor startup is time well spent. Of course, if money is your most important objective, then Web 2.0 and related startups which usually require significantly less startup capital might be a better option at this point. Although from my experience, many semiconductor aficionados will very much cringe at the idea of writing code all day or having to sit through code reviews.

A few weeks ago Nick Tredennick published a rather interesting analysis of the current state of the semiconductor market, titled Computing in Transition. Now why would you care about what Nick has to say about computing? Well, let’s say he has been around the block a few times. He worked on microcontrollers for Motorola in the late 70s, and then worked on the design of the Micro/370 microprocessor at IBM in the eighties before becoming the director of product development at Nexgen, followed by a quick stint as the chief scientists at Altera in the 90s. Currently he holds the president post at Tredennick Inc., a company specializing in consulting for full-custom and semi-custom VLSI designs. And if this were not enough, he also is an editor for the Glider Technology Report. As you see, when Nick publishes a presentation it is probably worth reviewing to get a viewpoint from an industry veteran.

In his latest presentation the premise is that the microprocessor in essence stalled innovation in logic design since it allowed programming to become a substitute for logic design. This held true as long as the design goal was defined by cost-performance. However, more recently with the introduction of power constraints the design goal has been modified to be cost-performance per watt, leading to multi-core chips. As such, the computing market is in a transitional phase, however, where exactly it is heading is rather hard to predict. What might help therefore is an analysis of where the industry has been and where it is now.

Here Nick digs up some very interesting facts and charts: For example, while it seems that companies are shipping millions of chips yearly, the semiconductor market accounts for less than one percent of the gross world product (GWP). At least the semiconductor market is growing at about twice the rate of the GWP. How is this for another interesting prediction: Unlike technology pundits who like to predict future killer applications that will take the market by storm, Nick has a rather simple prediction: there will be no such application. A couple other things get cleared up as well: First, Motorola destroyed Four-Phase Systems leaving Intel to dominate the microprocessor market and Altera is older than Xilinx. On the more serious side, Nick does a great job breaking down the current market into different segments, Microprocessors, ASICs, FPGA, etc. He then also dissects the market based on design goals: zero cost, zero power, zero delay, and zero volume.  These two approaches are then used in turn to analyze where some of the current products fall. But that’s not all, more analysis follows taking into consideration transistor scaling, yield implications, processor complexity and design effort. Eventually, Nick arrives at the prediction that reconfigurable hardware might well be the answer to optimizing future designs for the particular function they are to implement at a particular point in time. How exactly he gets to this conclusion is best understood by reading his presentation. One more thing, the presentation is almost 40 pages long, so you better grab a coffee and a comfortable chair.