The semiconductor industry has been through big changes over the past three decades, as suppliers continue to shed many of the functions that were de rigueur for chip makers in the '50s, '60s and '70s. These days, a semiconductor company no longer pulls its own silicon ingots, slices the wafers and prepares them for processing, creates its own masks, writes its own full suite of design automation tools and creates its own custom test systems. Most of those functions are now part of the industry infrastructure support.

Although many semiconductor companies have cast off those functions to reduce overhead and to focus their staffs on their key areas of expertise--typically, the architectural definition and physical implementation of the circuits--I wonder if they are starting to go too far. By reducing the variety of internal research and technology development efforts they undertake, these companies have fewer cross-fertilization opportunities and, thus, fewer opportunities to discuss unorthodox or even off-the-wall concepts that someone from outside the current project might bring to the design.

This hollowing-out of companies also puts a great burden on the external infrastructure to supply all the missing pieces. And the process has not stopped.

The fabless semiconductor vendor is a common business model today, with wafer foundries doing all the manufacturing. Such a company just has to design the chip, select a foundry and then work with that foundry to guide the chip through the manufacturing process. But even routine interaction with a foundry is being offloaded to service companies, which are sprouting up to manage the design through fabrication. As a result, the fabless company no longer has to have anyone on staff who knows how to manage the manufacturing process.

Many intellectual-property suppliers already offer a great deal of the key functions that will be integrated on a chip. In addition, many design support companies can design either a small part of the chip or the entire device, based on its architectural or functional description. This further ability to shed more functions reduces the fabless semiconductor company to a staff approaching zero--just a few system architects who define the unique aspects that differentiate the design and provide the "value-added" functionality.

Will those architects be the next group to be outsourced, hollowing out the company to just a management shell? If so, the remaining staff might consist of an "idea person" who then hires all the subcontractors to do the work. I fear that such a direction would be troublesome and prone to inefficiency, not to mention leaving many engineers looking for gainful employment. It could also slow down innovation, since there would be minimal interaction among the task groups and very little sharing of information or teamwork.

In that case, the days of sharing ideas around the proverbial water cooler will be gone.