An ASIC methodology emerges in Taiwan

In a recent conversation with Keh-Ching Huang, director of marketing at Global Unichip, the subject of ASIC business models and the closely-linked subject of ASIC methodology, came up. Global Unichip, if you are not familiar with them, is a fabless ASIC house closely associated with TSMC, in much the way that Faraday is associated with UMC. Huang said that the rules of engagement with customers have been changing of late, and that new forms of engagement are influencing not only the way Global does design work, but the degree of expertise the company must bring to the table.

In the good old days, Huang said, the model of engagement was often a netlist sign-off. The customer would bring in a netlist and simply ask Global to do the physical design and verification. But as the netlist sign-off has become more and more occluded by issues that used to stay politely in the back-end space, and as customers find they are lacking more and more of the IP necessary to do today’s system-level chips, the model is changing.

Huang said that much of Global’s work these days goes into gathering and validating IP-based platforms: libraries of functions, often in hard-macro form, that are necessary for a particular application space. If the functional IP is on the shelf, verified, outfitted for power management, are DfM-friendly, then creating a specific chip for a customer becomes mainly a matter of IP assembly rather than a physical design effort.

The background work necessary to do this is extensive, Huang said. Global attempts to acquire enough domain expertise to make correct choices about IP—what functions customers will need, and how to judge best-of-breed from the viewpoint of users as well as in the view of chip designers. Once the company selects a set of IP, they create a fully-featured hardware platform. This happens first as an FPGA prototype—later available to customers as a development board–then as a SiP, keeping the difficult bits like precision analog blocks on separate silicon. Finally, the company will tape out an SoC, so that in principle a superset of the platform actually exists in silicon before customers try to use it. In parallel to this hardware effort, Global builds a driver set for the chip. Application and user interface code the company leaves up to customers, as that is their primary means of differentiation, Huang says. The approach is strikingly similar to that recently described by Faraday, which is also based on a full-featured platform SoC which the company simply depopulates to meet the requirements of specific customers.

Global has used this approach in a number of consumer electronics segments, including a current effort on mobile video. It has also pursued a platform for VoIP. Even though Global has a close working relationship with TSMC, the company pursues IP independently from the giant foundry’s IP certification program. Of course Global will use TSMC-certified IP if it can, but often, Huang said, customers need something that hasn’t been around long enough to be certified, is too specialized, or simply comes from an IP vendor too small to catch the attention of giant TSMC. Sometimes, according to Huang, a customer need or a function Global discovers to be latent in an application area will lead Global to qualify an IP block, and then call TSMC’s attention to it.

The greatest struggle, Huang says, is for Global, whose background is after all in physical SoC design, to develop domain expertise in consumer, handset, and communications markets. This knowledge is necessary to be an informed platform builder, of course. But beyond that, according to Huang, customers are asking Global to shoulder more and more of the load of SoC design. "We are even getting requests to help with product definition," Huang said. As an example, he cited the latest trends in the video surveillance space, where intelligence that used to reside in security guards’ brains is migrating into camera SoCs, so that suspicious objects or patterns of activity can be spotted and flagged before the data ever gets transmitted to the central office. The concept is great, but it is carrying camera vendors way beyond their comfort zone. And so they turn to their SoC partner to offer a lot more than just a good back-end design. And in the process, they may be describing the business model for the future of the fabless ASIC business.