Jim Hogan gave the keynote at today’s EETimes Virtual SoC event and he presented a deep dive into the opportunities in semiconductor device development and the associated EDA opportunities from the perspective of an investor who makes his money building, buying, and selling companies. (Note: NOT chips and tools, companies.) As always, Hogan displayed an immense grasp of the forces that move and shake our industry.
He started by noting that systems companies (such as Microsoft, Google, Oracle, and Apple) are increasingly relying on SoCs to differentiate their products and to deliver value. However, make no mistake he said, software now sells hardware (the Jim Hogan way of saying “apps-driven”), which is an inversion of the traditional semiconductor value chain we’ve had for the last 50 years. Why is this good for the semiconductor industry? Because, as Hogan discussed later in his talk, SoC profit margins are 40% to 60% while the profit margins for “discrete” semiconductors is 10% to 20%–much lower. As a result of the system companies’ focus on SoCs as differentiators, a lot of the system value falls from the system companies to the semiconductor companies. However, what Hogan didn’t explicitly say is that the semiconductor companies have generally been less successful at fully realizing that value from their customers, with some notable exceptions in the PC processor business.
SoCs, noted Hogan, are evolving into SSoCs (Smart or System SoCs), a term introduced by research firm IDC. (See the previous EDA360 Insider blog: Smart Technology: Is the world as ready for smarter embedded devices as IDC’s Mario Morales thinks?) SSoCs are far more complex than SoCs, with more on-chip processors and a lot of associated software. As a result, SoC complexity is quickly increasing and this trend accelerates the evolution of the IP used to design the chips. If you take a broad view of IP, down to the leaf cells, you can see that the trend over the past four decades or so is for IP to evolve from transistors to gates to IP blocks and finally to subsystems (with associated software). The notion of IP subsystems is something I discussed extensively in the EDA360 Insider after Semico’s publication of a study of IP subsystems last year. (See “The last word on IP Subsystems from Semico: Where are they going?“)
The rise of software in this evolution is something not to be ignored. The real key is the OS, said Hogan, who noted that Apple has been very successful in using the same OS over many of its devices (iPhones, iPads, iPods) and has also been successful in making all of its products interoperate easily across processor architectures (several flavors of the ARM architecture, Intel’s x86 architecture, and formerly the PowerPC). Different processors but the same world view, he said.
This observation led to Hogan’s analysis of SSoC market drivers. SSoCs, he said surprisingly, are not processor centric. Then he explained what he meant. SSoCs are multi-CPU, multicore, multi-DSP, and have multiple hardware accelerators. There’s a lot of processing taking place, but no one processor or processor architecture is performing all of the processing. The identities of the on-chip processors are masked to a great extent, although ARM dominates the mobile market even with this anonymity said Hogan.
He then said that the announcement that Microsoft Windows would soon run on ARM processors would break the Wintel franchise. That in itself is a major market-changing force. “Intel will have trouble displacing ARM” in the mobile market, said Hogan. He therefore predicted that Intel will take an ARM license, probably an architectural license, if it really wants to break into the mobile arena. To reinforce that point, Hogan projected a semiconductor vendor ranking from early 2010.
This chart shows that of the top 10 semiconductor vendors at that time, four had ARM licenses. Hogan didn’t point out that three more are essentially memory companies and are not in the processor business and that TSMC is a foundry. That leaves Intel and Renesas as the two top-10 semiconductor vendors making processor-based chips that don’t have ARM licenses. Also note that most of the remaining semiconductor vendors in the top 20 list also have ARM licenses. Intel’s main competitor in the mobile space is actually Qualcomm, said Hogan.
None of these SSoCs will be designed without an evolution of the way IP is used, said Hogan. He then displayed this image showing that IP is quickly evolving from blocks to IP subsystems.
There are several key changes taking place here. One is that an IP subsystem consists of several IP hardware cores (possibly a controller plus a PHY) that have already been shown to work together. They are pre-integrated. On top of this, the IP subsystem also has firmware associated with it. So, for example, it now makes sense to license a USB subsystem because there’s no differentiation there. There’s no differentiation in designing the hardware and there’s no differentiation in developing bare-metal software to run that hardware. The differentiation is in what the system can do with that USB port and that differentiation is in the upper software layers, not in the bare-metal drivers. (Note: What’s not shown in Hogan’s illustration is the essential verification IP (VIP) needed to prove that the IP subsystem is properly integrated into the design. VIP is an essential IP subsystem component.)
Another big change that’s happening is in the SoC design flow. If software is truly coming to the fore (and there’s no one really arguing with that premise), then software development must occur sooner than ever. That’s leading to the following design flow:
Software developers need a software virtual prototype to start their development long before there’s hardware. In fact, they need this early start so that they can influence the hardware design if necessary–when isn’t it necessary?—and this is something that’s not occurred with any regularity in SoC design to date. Yet it’s absolutely required for true design optimization.
After this development phase, said Hogan, the implementation team needs to be able to construct a hardware virtual prototype to explore their architectural ideas. Hogan places the hardware virtual prototype in the SoC Realization category. As an investor interested in SoC Realization companies, he might be telegraphing his interests here.
Below the software and hardware prototypes you find the domain of Silicon Realization, what Hogan calls “EDA Classic.”
Hogan then gave a detailed picture of what he thinks SoC Realization looks like:
Cycle-accurate models sit on the boundary between System Realization and Silicon Realization. There’s a lot of opportunity there for an EDA investor like Hogan. In fact, he ended his presentation by putting up his “incomplete” shopping list of tool capabilities that he believes are needed for SoC Realization. Here it is:
(See the previous EDA360 Insider blog: “What’s on VC Jim Hogan’s SoC Realization shopping list?“)