If you're starting a new semiconductor company, the issue of building a semiconductor fab facility never comes up. You use one or more of the many excellent worldwide foundries to manufacture your chips. But what do you do if your 20-year-old company has its own fab? If you're a multibillion-dollar semiconductor giant, that's not a problem. If your revenues are in the $100 million range, though, you might have a serious problem.
Xicor faced this situation three years ago when I became its president. We needed to broaden our product line to improve revenues and cut operating expenses to im-prove profit margins. It was immediately clear that the captive fab was a major drain on our resources. I reasoned that a company of our size should invest its resources in IP, design, and new product development, not in manufacturing capacity or techniques. A captive fab violated that principle.
A captive fab can be a liability. It's a major resource drain. Also, it limits the range of products that can be manufactured, unless you're willing to increase the resource drain to upgrade the fab almost yearly. Plus, it limits manufacturing flexibility.
Simply maintaining a fab through maintenance and equipment replacement, even without technology upgrading, is costly. You must also keep a captive fab running at or near full production, because the cost of running at half capacity is almost as much as running at full capacity. In the "boom or bust" cycles of some markets (like memory that made up most of our revenue), it's virtually impossible to maintain full capacity consistently.
If your company follows market needs with its product lines, you eventually will have to upgrade your fab's capabilities. Doing that or increasing capacity levels stretches the resources of billion dollar companies. But for a $100 million company like Xicor, it's prohibitive. That limits the range of products that you can design and build, unless you supplement your fab with one from outside.
Moreover, a single fab limits manufacturing flexibility. Many semiconductor markets run in cycles where manufacturing volumes must be adjusted to meet market demands. In addition, customers don't want all of their critical parts made at one location. Recent earthquakes in Taiwan temporarily shut down several foundries and scared many IC customers. Now they want their parts available from two or more foundries so that a natural disaster won't shut off their supplies.
For small semiconductor suppliers, this means working with two or more foundries and spreading wafer fabrication among them. But, finding two or more acceptable foundry partners can be a problem. They must support your technologies, or have technologies that are similar to yours. Say you make EEPROMs, then you need a foundry with analog and digital capabilities, but not necessarily the latest deep-submicron technologies.
Outside foundry relationships must be established before closing an existing fab. Once that happens, closing or selling an existing fab eliminates a major profit and loss impact. This allows a company to concentrate its resources in IP, design, and new-product development.
While moving to a fabless business model has generally gone smoothly, it calls for a change in management style. With wafer production normally handled by independent companies thousands of miles away, production managers can no longer micromanage this aspect. Outside foundries usually allow visibility into their process. You can know where your wafers are anytime, but you have little control over the flow. In the three years that we've been using an outside foundry, the lack of control hasn't been a significant problem for us.