Posted by mike demler on January 18, 2008
If that popular game show ever has a category called EDA MYTHS, I expect this to be right up there for the Double Jeopardy round.
(Although, based on my experience in a failed startup… maybe Final Jeopardy?)
Remember, you must pose your answer in the form of a question…
Contestant> “I’ll take EDA MYTHS for $1000, Alex.”
Alex> For $1000, the answer is … Because digital design is so easy!
Contestant> “Why is analog design so hard?”
Alex> You are CORRECT!
I don’t know why, but for some reason over the past several months there seems to be a fixation in the EDA press on questions like “Why is analog design so difficult?”, or “Why is analog design black magic?” which I already commented on in one of my earlier blog posts back in October. You don’t see analog designers asking these questions, so why is this type of question so popular with the analog non-cognoscenti?
Analog non-cognoscenti: One who has never actually done an analog design and has no expert knowledge of the subject.
These questions are typically being posed to EDA industry marketers, who are also typically not experienced in analog design, to defend why “analog tools aren’t successful’” as Ed Sperling claimed in his recent DACezine article.
The problem, as I see it, is confusion caused by how easy digital design is. With digital synthesis you can create a working chip with no circuit knowledge; all you have to do is write some code. Why isn’t analog design this way? What’s wrong with all you EDA companies?
Or, since there have been several noteworthy failures of Analog Synthesis startups (mine was Antrim)… “Are Analog Designers Secret Luddites?; the question that Geoffrey James poses in his DACezine article?
I was interviewed for that last article, but was told that the topic was going to be “Way too Small: Will SoC and Mixed Signal be Incompatible?”
(My previous post here “Reports of the Death of AMS SoCs are Greatly Exaggerated”, puts the 1st question to bed because you can find numerous examples of 90nm, 65nm, and even 45nm AMS SoCs being presented at the upcoming ISSCC in San Francisco.)
Obviously, that’s a very different issue than the author’s allegation that it’s a “fact that EDA vendors have so far lagged in helping analog designers”.
(I addressed that accusation in my blog on Analog design is NOT black magic… but it is VERY hard as well).
But back to the analog synthesis question… what is synthesis anyway? Well, m-w.com says that synthesis is “the composition or combination of parts or elements so as to form a whole”. So let’s not forget that somewhere, someone created a circuit for all those logic gates that populate a cell library in the digital synthesis flow. And the truth is that digital design hasn’t required circuit design expertise for a long… long time, going back to at least 7400 TTL logic days and probably even earlier. Some circuit designer built those chips too, so that the digital designers could “synthesize” their PCB.
I think that the problem with all the NC (non-cognoscenti) thinking is this:
There is no Boolean Algebra for Analog!
Digital design is by definition easier than analog design, now there is a FACT. You can perform digital design with Boolean logic, simple rules and tables like Karnaugh maps. I’m sorry, but analog design does not work that way and it never will!
However, there is in fact an analog synthesis process that is performed in hundreds of companies every day, and guess what… it’s enabled by some very powerful EDA tools. The EDA industry has not “lagged in helping analog designers”. That’s just nonsense!
The EDA industry has facilitated orders of magnitude increases in analog designer productivity. If that wasn’t true there would be no consumer electronics industry today. Nobody could afford the time it would have taken twenty years ago to design the mixed-signal devices at the heart of today’s cell-phones and iPods.
I think it was very interesting to read, in Ed Sperling’s DACezine article, that “the biggest problem (at National Semiconductor) is simulation”. When I first started doing analog design, my synthesis process lacked the ability to simulate complete circuits like PLLs, and ADCs. Today’s SPICE and Fast-SPICE tools can handle these types of circuits routinely, and in a small fraction of the time it would have taken back then. Simulate a circuit with tens of thousands of transistors? (or maybe more)… you’d have to be nuts to try that twenty years ago. But, just as Design Compiler is now inextricably a key to the digital synthesis process, SPICE and Fast-SPICE is at the very core of the analog synthesis process.
And designs get bigger and faster, and more issues come with each new generation of nanometer processes. So we must constantly make our simulators go faster while analyzing more complicated device models efficiently, increasing capacity for more parasitics and layout effects, in order to keep up with designer’s requirements. It’s the same for digital as it is for analog. Bigger. Faster. (Dare I say cheaper.. ouch!).
So my final word (today) to the NC folks is that EDA has facilitated analog synthesis. It’s just not the same as digital synthesis. It shouldn’t be, and it never will be.
Let’s focus on the real problems, rather than propose solutions in search of problems. The EDA industry hasn’t failed to help analog designers. Custom analog design is not a “plague”. Analog designers are doing amazing things, and analog EDA tools are a key part of that.
I hope to see many of you at ISSCC to hear about all the latest mind-boggling transistor-level innovations, from the minds of the analog cognoscenti!