IP, SoC design, and the system OEM: changing roles and a Microsoft view of the future

A panel of system OEMs and fabless ASSP companies explored the changing relationships between chip vendors, IP providers, and the system OEMs who increasingly depend upon them both at the ARM Developers Conference in Santa Clara this afternoon. Surprisingly, there was little consensus on a single direction in which the industry is moving. Rather the panel became an exploration of diversity.

There was general agreement that the adage of the bubble era, that real OEMs have COT chip design teams, had passed into history. Today, as Hewlett Packard senior engineer/scientist Mobashar Yazdani explained, the whole structure of the system design chain is created with a case by case analysis. If an existing ASSP will do the OEM’s job, great. If a substantial modification to an existing design is necessary, the OEM may still choose to work with a fabless chip vendor. But if the OEM doesn’t see anything out there that fits their needs, Yazdani said, HP for one will not hesitate to flex its internal chip design muscle and work directly with a foundry.

The input that has spread out this spectrum of options is a seeming dilemma. On the one hand, extreme cost pressures, particularly in consumer markets, have made OEMs sensitive to every penny of BoM cost. This in turn has led them to drive down SoC and ASSP prices to the point of vanishing margins for the fabless semiconductor houses. “We have come to the point that we will simply walk away from some business,” said PMC-Sierra COO & GM Colin Harris.

But at the same time, if OEMs succeed in commoditizing the SoCs they depend upon, they undermine their own ability to differentiate. Software by itself often can’t carry the burden of product differentiation at reasonable performance levels.

This has led to a number of fascinating variations on a theme. In many designs, panelists indicated, the OEM enters a joint development with their silicon supplier so intimate that the OEM’s system designers reach clear through the fabless company to work with IP suppliers and even, according to Microsoft principle program manager Rob Rossi, to influence process decisions at the fabless company’s foundry. It is a picture of enormously powerful OEMs reaching back to create differentiation as far back in the design chain as necessary, crossing boundaries of company, technology, or—some would argue—expertise.

At the same time, Conexant chief development officer Anil Mankar said, many OEMs with less technology power are doing just the opposite: pushing all the technology decisions, including most of the software responsibilities, onto their SoC suppliers, so that the OEM becomes almost more of a value-added reseller than a system integrator.

But is there a trend? The consensus of the panel, under questioning from Gartner research vice president Bryan Lewis and an alert audience,  seemed to be that there is not. Individual OEMs will continue to make decisions based on their technical expertise and their view of end-product differentiation. The most powerful OEMs will continue to evaluate each market opportunity on a case-by-case basis, employing different kinds of relationships in different markets.

As an example, Rossi provided one possible future in the view of Microsoft. A differentiating technology would be developed in Microsoft Labs as a pure software implementation of an algorithm, Rossi said. In fact, he pointed out, many such algorithms are in Microsoft’s hands today, with potential to disrupt some apparently maturing consumer markets such as the digital still-camera market.

This software implementation would be moved from the server world onto a fairly general embedded heterogeneous multiprocessing platform with enough power and compactness to provide a compelling market entry. This would normally involve partitioning the software into a number of relatively independent modules to execute on separate processors in the platform. Then as the market developed, Microsoft would work with the provider of the platform SoC to reimplement specific modules either in software on accelerated processing cores or in dedicated hardware cores, all within the platform architecture. This would quickly drive the product down the cost curve and up the performance curve simultaneously, without necessarily relying on a new process node to reduce die size. The resulting product life profile, Rossi suggested, fits perfectly into the sort of media-based consumer markets in which Microsoft can extract value not only from the physical device but also from the content—the licensed video, music or whatever—that the user consumes.

Thus the panel painted the OEM/SoC relationship not as a trend, but as a kaleidoscope of options, relationships, and capabilities, determined by the market power, the technical expertise and the creativity of the OEM, and by the flexibility of the fabless SoC vendor. It is certainly a new world out there.