The SDE Circle of Life

There seems to be a big picture phenomenon that I like to call the System Design Enablement (SDE) Circle of Life, whereby IC design knowledge switches from being something to be outsourced to be the only route to differentiation, and back.

If you have kids, then you've probably watched The Lion King, perhaps several times. Also, the stage show version is a completely different experience, wonderful for any age. The opening song is Circle of Life. The lyrics are by Tim Rice, and music by Elton John. Apparently, Elton John wrote Circle of Life in two hours, including recording a demo—pretty quick for an academy award nomination for best song! (It didn't win, though; it lost to Can You Feel the Love Tonight by...Tim Rice and Elton John, also from The Lion King.) If you want to see him perform this trick live, here he is setting the instructions for actor Richard E. Grant's oven to music—on the spot!

The Seventies: Not Just Bad Hair

In the primeval days of semiconductor design in the 1970s, all the knowledge for how to do semiconductor design resided in the semiconductor companies themselves. It was the era of what Carver Mead called the "tall thin man" who needed to have a working knowledge of everything from process to layout to transistor-level design (there was no other kind yet). Since an integrated circuit could only hold a few gates, system-level design was done by putting a lot of them together on a board. This was the era of 74xx TTL design (I still remember that a 7474 contained 2 D-flipflops, and a 7404 contained 6 inverters). There was no system knowledge in the semiconductor companies, the building blocks of the 74xx chips were completely generic. In that era, semiconductor design was a black art that was not understood at all in the system companies, they just purchased building blocks and put them together. In this era, Lego really was a good analogy. The Lego company just builds the blocks without caring what kids build (at least before the era of pirate ships and Millennium falcons). Kids make no pretense to understanding the intricacies of plastic injection molding. This era was the start of the SDE circle of life.

Application Specific Integrated Circuits (ASIC)

In 1979, Mead & Conway came out. It was actually called Introduction to VLSI Systems, but nobody ever called it anything other than Mead & Conway. I consider it the book that changed everything. Looking back, the key thing was that it took semiconductor design out of the semiconductor companies and into the universities, leading to a generation of engineers (of which I am one), who understood how to design. At the same time, Moore's Law was in full swing, and so it was no longer attractive to do all design using little chips that only contained a couple of flops. A chip could hold 10,000 transistors. Some semiconductor companies came up with their own ideas for more complex building blocks to use the transistors: the first microprocessors, UARTs, CRT display controllers and so on. These were largely targeted at the nascent market for personal computers (the Apple II had been introduced in 1977, the IBM PC would be introduced in 1981).

However, some system companies wanted to build their own systems that couldn't easily be built out of these standard components. They wanted to do their own designs. They had the system knowledge—but not enough semiconductor knowledge. The semiconductor companies had that but didn't have the system knowledge. Two companies, in particular, stepped up to fill the gap: LSI Logic and VLSI Technology (which I joined). They invented what came to be known as the ASIC methodology. The system company would design at the schematic level using tools from the early EDA industry (primarily Daisy, Mentor and Valid, although VLSI had its own tools, too). They would simulate the circuit's behavior using gate- or transistor-level simulation and estimates of capacitance. The result of this design process was a netlist that would be sent off to the ASIC company, who would do the physical design. Circuit extraction would calculate the parasitic capacitance (nobody worried about resistance yet) and the values sent back to the system company, a process known as back-annotation, and the system company could re-run the simulations to make sure that they still passed with real values in place of the estimates. This was the next phase of the circle of life, system design being done in the system company, semiconductor design being done in the semiconductor companies.

Customer-Owned Tooling

The EDA industry made standard-cell place-and-route a more routine task, and they (and specialized companies like Artisan) started to supply standard cell libraries and memory compilers. At the same time, foundry capacity became available. This was either excess capacity from existing semiconductor companies, or else newly created specialized companies, so-called "pure-play" foundries. Jerry Sanders became famous for saying "real men have fabs", but he was on the wrong side of history. This combination—physical design tools, merchant libraries, and foundries—meant that system companies didn't need the semiconductor companies for anything other than manufacturing. They could design their own chips from system down to silicon and then contract out to have it manufactured. For technical reasons I won't go into today, this was known as COT (which stood for customer-owned tooling, don't ask). This phase of the SDE circle of life had the system companies doing everything but the manufacture.

Back to the Semiconductor Companies

The mobile handset industry, which in those days was led by Motorola, Nokia, and Ericsson (remember them?), was the first to realize that they were spending a lot of money on designing chips, that each process node got harder, and that maybe they should leave it to the experts. Microprocessors were available as IP—in particular from ARM—so the system companies could focus on the software and push semiconductor design off to companies like TI, VLSI Technology, and ST Microelectronics. So they did. This was the next phase of the SDE circle of life. System companies went back to doing the systems, only the software, in many cases, and the semiconductor companies designed the chips (either as standard products or as ASIC designs to a specification from the handset manufacturer). The next phase of the SDE circle of life had everyone going back to their roots again, with system companies doing just the system (and the software, which was a huge part of it) and the semis doing all of the design and manufacturing.

To be continued on Monday...