Is chipmaking skill still a differentiator?
09/01/2001
 |
Winning isn't easy in the tough, rapidly evolving semiconductor market. Many of the new applications call for specialized chips to maximize performance and minimize space and power requirements. Designers want lower power, faster processors, less latency for stored data, more functions on a chip (often including processor or DSP/memory/programmable logic and even analog), quicker graphics handling, and a widening array of other tough, often conflicting requirements for chip fabrication. Vendors want high-resolution color displays in wireless handsets for surfing the Web, along with text handling for e-mail and messaging. Game designers want super-realistic, fast-paced animation with 3-D graphics and perhaps some artificial intelligence to tailor responses to users. Notebook PCs run multiple programs concurrently for high-powered professionals and then, to ease the tension of a tough day, may run movies or music videos on high-resolution screens from a plug-in cartridge.
To meet ever-tougher demands, the chipmakers have been forging ahead of the Roadmap, beating rather than tracking Moore's Law. This isn't easy, because new materials are needed. Devices don't scale so easily when stray capacitance or leakage currents erode critical parameters. Dipping to 0.13µm (and maybe beyond) using 248nm steppers requires a lot of reticle enhancement, and the techniques and software to make this convenient are still under development. Putting different functions onto the same chip may require mixed processes, for logic, memory, and analog. Yet chipmakers have been meeting these growing challenges by a combination of clever design, careful analysis, and process wizardry, often with new materials and processes.
Looking at the history of the industry, it would seem that chipmaking proficiency would be rising in importance as a vital differentiator in the marketplace. The vendor that can make the highest-performance chip at the highest yield on the smallest die, and do it first, should sweep away the competition. In past times, it would have taken one of the most adept integrated device companies to achieve the toughest goals. Yet many of the hottest designs are coming from fabless companies, with advanced chip designs that can boost performance, lower power, and add features — and they are cranked out on shortened design cycles.
They can do this because the foundries have pushed their capabilities to match or even exceed the IDMs. They have also shortened design and production cycles to help fabless companies get their exotic designs to market faster.
There have been rumors of some of the IDMs giving up on chips altogether to concentrate on system design and marketing. Siemens spun off Infineon, TI sold off its DRAM fabs, and Motorola has been said to be considering dropping chipmaking to concentrate on its other businesses.
But the view that manufacturing prowess is no longer a key differentiator is not universal. Intel made a strong statement by announcing an increase in capital spending even in a down year. Much of that spending will be for advanced process tools to push chipmaking capabilities. While DEC never made a success of its Alpha microprocessor, its process team at the fab in Hudson, MA, pushed toolmakers to advance the state-of-the-art to keep the Alpha's capabilities out front. Samsung learned from DEC, and now continues to advance Alpha capabilities as the Korean company moves beyond being just a leading DRAM maker. Japanese companies that are fierce competitors in some markets are joining forces in chipmaking so that they can maintain strong fab capabilities, especially for system-LSIs, even as their fortunes ebb and flow.
IBM may be the best example of a company dedicated to leadership in chipmaking. In earlier days the company made sure it stayed ahead in chip fabrication so that it could offer the most powerful mainframe computers. In recent times, Lou Gerstner has focused the company on networking and networked computing, with its chips as well as systems, software, and services. Innovations like copper, SiGe, and SOI came into the market early from IBM. The company also has pushed its process capabilities, such as its development of CMP and early adoption of minienvironments, and it was a pioneer in advanced packaging. IBM doesn't farm out its advanced chips to foundries, instead it provides foundry capacity for some of the most advanced chip projects of other companies. Its R&D can move rapidly from laboratories out into the fabs, for new types of devices such as ferroelectric memories, as well as process advances.
Which companies are right? Note that a linkage is taking shape between the early chip design phase and intricate process steps in the fabrication cycle. This is being pushed by the need to insert reticle enhancement techniques early in the design flow, so that their effects can be dealt with before troublesome parasitics show up on silicon. Better models will be needed to make this process work well. But the exercise will teach chipmakers that early tradeoffs can be made between tweaks of the design and later process steps. As the slop disappears from design rules, and process latitude shrinks, the most advanced chips will require this kind of early design tweaking. That will be hard to do with a foundry, which may not want to share the nuances and weaknesses of its process technology with outsiders.
Robert Haavind
Editor in Chief