Inspired by her mother's ghost, a girl friend of mine kept pumping $20 bills into a Reno slot machine. Her luck was pretty good at first: She won $18 and then $6, and then hit some sort of $250-kitty, and then another on the same machine. People began to gather round, drawn to the "ding-ding-ding-ding" sounds the machine was making, astounded at her good luck. On a nickel-bet machine (maximum bet for a quarter) and an investment of $40, she won over $500! No: it wasn't the $20,000 jackpot the flashing lights above the slot machine kept promising, but a pretty hefty return nonetheless.
"Cash out!" I kept telling my girl friend.
"No, I can't stop now," she refused, getting a zero return on one bet, and zero again on the next. People began to drift away. "My mother is looking out for me," she claimed.
"Cash in your $500," I suggested. "Put $400 away for safekeeping. Play some more with the other $100, and when that's gone, quit." My logic was admittedly NOT the gambler's, the romantic movie hero who risks everything on a throw of the dice. It was more of the family man's, the guy who lives perhaps drearily from paycheck-to-paycheck, and attempts to stay within a budget.
I might make someone a good husband. But in Reno, I was a lousy date. After losing $100, I was ready to call it a night. But my girl friend kept hitting the "Bet the MAX" button, oblivious to what I was trying to tell her. I went upstairs to read a book, and my friend inevitably pumped all her winnings and more back into a machine that was not the least bit maternal. She was ultimately consumed by the ghost of her parent in the same way that Hamlet was consumed by the ghost of HIS
I wonder sometimes whether we in the semiconductor industry aren't functioning as if we're in some kind of cosmic gambling casino. We're better at marketing now than we once were (that is, we know how to promote), but we can't forecast worth a dunghill. We're still functioning with the same "field of dreams" vision that inspired the development of 32-bit processors: "If you build it, they will come."
We heard Strategies Analytics project earlier this year that there'll be something like 670 million cell phones shipped this year — some of it reflecting the replacement of existing handsets with camera phone models; some of it reflecting still untapped demand in South Asia (India, China, maybe Pakistan). I think market research firms are entitled to make money selling their fantasies the same way science fiction writers sell THEIRS, but I'm not sure if I'd build factories and hire workers on the basis of these projections. I worry that we in the semiconductor industry are still churning out products as if it were not the calculated 670-million number we were trying to hit, but more like a billion — never for a moment asking whether there's an infrastructure in South Asia that could support that kind of consumption.
From the chair I sit in, the industry — including Intel, which sees ITS best growth path in communications — is still dancing to tune whistled by the cell phone makers. Against 23-26-week lead times, cell phone manufacturers had built inventory in 1999 and 2000. But somewhere in that period it became clear, they had too much product on hand, and abruptly told the semiconductor industry, "Stop!" We hit the excess inventory wall in 2000 — and that's when cell phone projections were on the order of 435 million units (not 670 million). Signs are all over the place that we're hitting the wall again.
It took us one and a half years to burn off the inventory we built in 2000. As the inventory was depleted, cell phone manufacturers — both OEMs and assemblers — started increasing their order rates. Some semi suppliers, looking at the quarter-by-quarter upward trend in orders, began declaring an economic turnaround ("the recession is over") when in fact it might have been just "inventory replacement." (Linear Technology Corp., in fact, was one of the first to declare a turnaround, opening job requisitions for engineering talent when others, like John Daane of Altera were still stroking their chins about how much faith to put in the upward trends they were seeing.)
The devastating recession sent a shock wave through the industry, and we have yet to deal with all the repercussions. We've sold off marginal businesses units, in the past four years, and laid off thousands of workers — including some very good engineering talent. We've sent guest workers back to their home countries, along with "outsourced" jobs. But we haven't completely "right-sized." Like the casino gambler whose lucky streak has abruptly slowed, we haven't thought through the steps for an end game, an exit strategy. We've capped some investments, but we haven't quite identified if there's limitations on how much we should continue to spend.
Somebody up there loves us
And why should we? This is the industry that the Clinton administration claimed employed more American workers than the auto industry. An industry Al Gore claimed built the "V8 engines for the Internet superhighway." (I was there when he said that.) This is the industry that showed 1200 and 2000 percent gains in stock price valuations; an industry that transformed the communications culture of all the world's industrial nations with its building blocks for computers, cell phones, and Internet access devices. This is the industry that now can't quite come to grips with the fact that it may be overproducing — designing and manufacturing products that nobody particularly needs or wants.
This assessment may be unnecessarily harsh and alarmist for someone in the industry. (It's definitely "unromantic.") But with the entire semiconductor industry tilted toward risky system-on-chip (SoC) projects utilizing the smallest (and most costly) CMOS geometries, and Wall Street favoring players with huge fabrication facilities — and continually monitoring their utililization — I'm wondering whether we're hitting a choke point.
The breakeven point for a CMOS SoC used to be something like 10 million pieces. But with projected NRE costs for a 90nm device in the $10-million region, the breakeven point goes out to around 100-million pieces. Who, outside the cell phone industry, is projecting that kind of volume?
But there's nothing on the horizon that replaces cell phones as the next big thing. Worldwide PC sales are still in the 60-90 million range. (Congratulations to Apple Computer for making the iPod music player the flagship of its new business, but its quarterly success story is factored on the sale of 2 million units — not the 100 million pieces cell phone makers need to make a good showing.)
Where does analog fit in?
Analyst Doug Freedman of American Technology Research has been talking up Linear Technology, Maxim and Analog Devices — even though these companies have been experiencing the same slow down in order rates (the same backups through the supply chain) as other parts suppliers. LTC has very good inventory management, he claims, and won't get caught off guard the way the rest of the industry was in 2000. (National Semiconductor has been learning this lesson as well, I thought I heard him say.)
But I also heard him say that LTC's premium pricing policy has slowly been eroding. The company's ASPs, once "North of $2," are now down around $1.40. There certain parts, Freedman indicated, which sell well below $1 — in 250-million piece lots. The question I have (apart from the shocking revelation about LTC's pricing trend) is whether the company can make any money on that part, if they don't sell 250 million of them? And, who, besides the cell phone makers, will use 250 million pieces?
I'm wondering though whether the analog companies — with their concentration on multi-market building blocks — might have a model for the rest of the semiconductor industry. Certainly there are new applications emerging in automotive and medical monitoring realms that will require a different set of design skills and manufacturing talents than we've seen with seen with cell phones and wireless whatnots. The SoCs going into these as yet-undefined applications will require higher breakdown voltages, but with higher sensitivity and lowered noise floors. But unless the auto industry, to pick a potentially good customer, goes into overproduction, the volumes for many part types will be considerably less than 100 million pieces. (In these places, a little company like Anadigm could make a profit selling Mercedes-Benz, say, 20,000 pieces of a programmable-analog driver circuit for motorized seats.)
This is a different model than "BC2BC" — big company-to-big company — that the biggest semiconductor makers are currently using to underwrite their manufacturing investments. It may call for, if not the "downsizing" of the semiconductor industry, a re-organization or a de-centralization, around higher-ASP parts which make the breakeven point closer to the 10 million pieces of yesterday, than the 100 million we worry about today.
In the meantime, the choices we as an industry face are easily as dramatic as political choices between free enterprise and socialism. I absolutely agree with the sentiment that says, "Let the market decide!" But are we alert and responsive enough to recognize what the market is telling us? Or do we need some kind of higher authority — a trade group like the Semiconductor Industries Association (SIA) or the government itself — to hit us over the head with a 2-by-4?
In this time of economic and political turmoil, we certainly do ourselves no good by producing devices which only contribute to inventory glut. At the very least, coming back to the gambling casino analogy, we might stop pumping our moneys into the same slots, even when they've been good to us in the past.
Your comments please: firstname.lastname@example.org