For Pasquale Pistorio, president and CEO of STMicroelectronics, the issue was quite straightforward: "It's the natural evolution of the industry," he said, "you can't deal with individual transistors any more, now you have to deal with blocks - blocks of IP [intellectual property]. Time-to-volume is only achievable by this means. By putting more functions together faster, you drive the technology towards negative inflation because you drive the cost per function down."
Jurgen Knorr, chairman of the pan-European R&D programme MEDEA, added that it wasn't just a case of integrating transistor blocks. "You can also put mechanical parts on a chip like the accelerometers used in airbag controls."
But increased levels of integration bring their own problems, as Fred Shlapak, president and CEO of Motorola Semiconductors, pointed out: "0.13µm technology is delivering 200,000 to 250,000 gates per square millimetre of silicon, and technology promises to increase that to 350,000-400,000 gates per sq mm in 18 months time - the issues of simulation and verification are going to be fundamental."
For Ulrich Schumacher, president and CEO of Infineon Technologies, the issues of SoC were: "We need better relationships with customers, we need systems know-how and we need software".
The latter was something with which Pistorio agreed: "In terms of effort - in terms of man hours - nowadays we spend as much time on software as on hardware".
However, the SoC business is not all gravy. "It's a binary business - one with big successes and big failures", said Arthur van der Poel, president and CEO of Philips Semiconductors. "If you satisfy all a customer's requirements, you get all of his business, if there's one small thing wrong in the corner, you get zero."
Pistorio listed the ingredients for success in SoC: "You need systems knowledge; close relationships with customers; a wide variety of engines - DSPs, microprocessors, microcontrollers etc; a broad IP portfolio; a broad product portfolio; design technology; a manufacturing machine capable of producing deliverables."
Asked if manufacturing wasn't now something that could be subcontracted to a foundry, he replied: "I don't believe so. If you want to embed DRAM, flash, sensors etc, you need your own manufacturing capability. In the SoC business, a fabless company can only attack niche markets."
Shlapak agreed: "In GSM, for instance, it's impossible to use a foundry and make a buck."
Schumacher agreed up to a point: "You have to have the ability to integrate many engines, but you don't have to own them all, it's the same with IP blocks," he said, "foundries are good for people who provide part of a system. But for high level integration, you also need assembly technology."
Van der Poel added: "You may not have time or the resources to develop IP on your own. You have to focus on a few segments but customers expect you to have everything. We're replacing NIH [not invented here] syndrome with PFE [proudly found elsewhere]."
There are, however, various levels of IP. "Strategic IP, which gives competitive advantage, is something which you develop yourself and keep to yourself. It is not found in a supermarket," said Pistorio. "Commodity IP is something you can find in a supermarket but gives you no competitive advantage. The semiconductor industry has moved into a brain-intensive era - the biggest challenge is to find the brains when we have hundreds of thousands fewer engineers than we need - and to organise them to work productively."
Both Van der Poel and Pistorio agreed they were spending two-thirds of their R&D budgets on SoC. However Pistorio added that: "If you define an SoC as an engine with IPs around it, then SoC accounts for one third of the total product revenues of ST."
So is the SoC business basically a custom product sector? Knorr and Shlapak thought not. "When you have only a few functions on a chip, you need many chips - the lower the complexity the more suppliers you have", said Knorr, "the higher the complexity, the more it's dedicated to one guy and the volume per chip will decrease. So the question is how to develop platforms with re-usable IP which can be changed with software to give a specific capability to your customer at the lowest cost."
Shlapak took a similar line: "SoC is not a custom business because you produce software-defined platforms - platforms which produce products three to four months after you've produced the platform".
But Pistorio reckoned: "We do universal sockets for some customers, but mostly each company wants its own solution. The larger a customer is, the more he wants his own solution."
Second sourcing is not something which happens in the SoC business. "Customers think about time-to-market and time-to-money", said Shlapak, "he hasn't got time to get a second source. He's locked into one supplier."
Pistorio agreed: "In any given socket, a customer has to work with one supplier only. So the customer has to be very careful in his selection of a supplier. And there are fewer and fewer suppliers on which a customer can rely."
Does this mean profitability in SoC is high? "No," said Shlapak, "customers want a competitive solution. Customers are very well educated - they want a value proposition in front of them, and they have a big filter."