Recent press reports have indicated that the semiconductor cycle has either died or will be muted from now on. That's a tempting prediction to make if a downturn/upturn doesn't materialize on schedule or when you're in the middle of a muted cycle. One factor being credited for this phenomenon is improved inventory management.

Today's downturn is indeed muted, so Semico is not surprised to see such predictions being made. We argue, though, that in comparison with the 2000 Internet demand surge, and the subsequent 2001 collapse, almost any cycle would seem muted! Add to this the large size of today's NAND market, and its current counter-cyclical behavior, and the result is 2004's soft upturn and today's soft downturn.

Semico believes that in a few years we will stop seeing the kind of DRAM-to-NAND fab conversions that are being performed today, and when these end then NAND will end its counter-cyclical behavior, going into oversupplies and undersupplies simultaneously with the rest of the semiconductor market. When that occurs, the severity of overall semiconductor cycles should approach to the more dramatic levels they attained in the past. In the mean time cycles will continue to be atypical!

There are higher-level reasons why we are seeing less dramatic silicon cycles than we have in the past. One key reason is that we no longer have such strong demand growth as before. The industry rode Moore's Law for processors while we extended IBM mainframe architectures down to the desktop, to the point that Intel seems to have admitted that the raw megahertz of processors has exceeded the requirements of most high-volume applications. The industry drove the density of semiconductor memory up and the cost per bit of memory down to the point that companies are using solid state memory to replace magnetic storage. One result is that the industry hasn't done a full 4x generation step for DRAMs since the 16Mb-64Mb conversion back in the late 90's. The semiconductor industry is basically in idle mode while trying to find the next single-focus technology driver. Yet, capacity continues to be added.

Our current analysis of capital spending suggests that even if companies make a concerted effort to constrain fab growth the industry will see utilization rates fall again in late 2006 as the capacity started in 2004-5 enters volume production. [Editor's note: recent DRAM and NAND price declines suggest over-supply could strike much sooner.] DRAM and NAND companies build fabs and equip them to ensure they have lots of capacity with little consideration of the "big picture" vision of future world supply. That is, they assume that they'll increase percent market share or they assume that worldwide semiconductor demand will be much greater than Semico forecasts in our wafer demand model.

Let's return then to those predictions of the demise of the semiconductor cycle. We have seen several predictions in the past crediting inventory management improvements as one of the key reasons silicon cycles will die off, so we take a very jaded view of that argument. Even though there is more process automation in the channel, several other factors have increased in complexity: the number of links between the actual silicon manufacturer and the final user of that silicon have increased,  the OEM product options and variations have increased, and the levels of specialized services have increased (such as assembly, subassembly, distributors, outsourcing, contractors, contract manufacturers, subcontractors, etc), while the semiconductor manufacturing cycle remains between 2-3 months. [Editor's note: Another vote for taking recent claims of the end to excess inventory with a grain of salt]

In the end it is unlikely that the supply chain as a whole has really improved all that much. "Better inventory management" is the euphemism for "the semi manufacturer eats more of the excess inventory caused by poor forecasting because we are in a long cycle of over capacity." Today's biggest improvement in inventory management stems from the fact that the excess material is scrapped by the semiconductor company at earlier stages than before.

The end of the semiconductor cycle? Not quite!