Automating Analog Design

Is it really a black art or just a red herring?

By Geoffrey James

For years, the EDA industry has been calling analog IC design a black art–tool makers wag their heads sadly at the difficulties of automating it.  Many industry experts still espouse that view, but there are a growing number of voices saying that the challenge has been overrated and that the recalcitrance of the analog designer community might be as much of a problem as the inherent complexity of analog design itself.  Is analog design truly a black art?  And, if not, why do so many think it is?  And when will the EDA industry see the breakthroughs that will automate analog IC design, thus transforming it from an art into a science?

State of the Black Art
Nobody seriously questions that designing analog IC circuitry (including mixed-signal) is more difficult than designing comparable digital IC circuitry.  Unlike digital designers, analog circuit designers must make 'architectural' decisions that balance a complicated set of concerns, according to Eric Filseth, CEO of Ciranova, an EDA startup that's building tools for analog designers working on custom chips.  As such, analog design has a lot of the same characteristics of computer architecture design, with many trade-offs and up-front decision-making.  "[It's] not very automated, requires experienced and insightful designers," Filseth explains.

Analog IC design also requires a deep understanding of physics and how components react to each other in the silicon environment.  As a consequence, "the study of analog design is very hard…much harder than digital design," says Mike Demler, product marketing manager for Synopsys's analog simulation tool. Analog design requires expert knowledge of the fundamental principles of electronics, circuit theory, semiconductor physics, signal processing theory, advanced mathematics and so forth.  As a result, "engineers who choose to pursue analog design as a career must study (and master) a large body of science," says Demler, "Those who can do that and then apply their knowledge creatively are scarce."

To make matters worse, analog design gets more difficult as the industry moves to ever-smaller nodes.  "The physical effects of nanometer processes have introduced variability, the need to manage leakage and power while maintaining performance with lower supply voltages, and the requirement for analysis of numerous post-layout effects," says Demler.  Furthermore, the complexity of design rules at 65 and 45nm and below makes it hard for human engineers to optimize an analog design (especially the physical layout) in an acceptable timeframe, according to Filseth.

Despite the complexity, though, it's probably an overstatement to call analog IC design a black art.  "There are definitely parts of the custom analog design process that are sufficiently rote to be computerized," claims Steve Lewis, a director of product marketing at Cadence. "The trick is to concentrate on removing some of the drudgery from the process," he says.  EDA pundit Gary Smith of Gary Smith EDA believes that at least some of the difficulties associated with automating analog design lie in a general unwillingness, among analog designers, to cooperate with EDA tool builders. "Analog engineers have been by far the highest paid of the two and analog IC vendors also have far better margins than the digital vendors," he says, "Keeping the perception of analog design as a black art has been in their best interests."

Analog Wagging the Digital Dog
Part of the problem is that, when it comes to engineers, "automation" is a loaded term.  While "automation" can be taken to mean "removing the drudgery," there's a dark side, too.  Analog designers can't have helped but notice that there have been entire classes of professionals in the high tech world who have found themselves "automated" out of a high-paying job.

Commercial software development is a case in point. Prior to the mid-1980s, programming was itself considered something of a black art, requiring plenty of hands-on communication and in-house tweaking.  Today, however, commercial software development is now so "automated" that much of it is farmed out to semi-skilled programmers in third world countries.

A similar "de-skilling" process took place among digital IC designers, according to Chris Collins, director of analog technology infrastructure at Texas Instruments.  "The best thing to ever happen to digital was when a voltage was quantized into a zero and a one," he explained, "Many great EDA tools were then developed around that sort of structure and it certainly made digital design into a commoditized trade...literally anyone could do it."  As a result, digital IC designers now command significantly lower salaries than their analog brethren.

Unfortunately, "automating" something as complicated as analog design is only possible with the direct cooperation of the folk being "automated."  And that's created a conflict of interest for EDA firms who haven't been willing to irritate the analog designers, lest they stop buying their tools. "If you annoy your constituency, they'll just say 'we won't do it," says Cadence's Lewis, "Most companies have a shortage of analog master designers and they're like the tip of the tail that wags the dog."

However, that dysfunctional dynamic is beginning to change as the result of two major trends, according to Gary Smith.  The first trend is the aging of the workforce. "The good news is the old farts are retiring and the new analog guys are willing to use tools," he says.  The second trend is the injection of foreign design firms into the mix. "Asia has come up as a major analog area and in Asia there is no separate analog/digital engineer categories; they are all just IC designers," says Smith.  And that makes them more willing to cooperate when it comes to co-developing tools that might add more automation into the analog design process.

This is not to say that there's will be a change overnight. "It is in my opinion that analog design is still a guild art in many ways and the best analog designers are Master Craftsmen in their trade," says Collins, who believes that it may be a decade, or more, before the promise of analog automation can truly reach fruition.  "The EDA industry is still very much today where it was in the late 80's when it comes to analog software innovation," he explains, "By and large, the analog EDA industry is still the 'netlist and simulate' EDA industry (which) continues to add to the mystique and black art persona of analog design."

Potential Breakthroughs
Even with the cooperation from analog designers, a major breakthrough is unlikely, according to Jeff Lewis, vice president of Marketing  at Innovative Silicon, a startup pioneer of "floating-body" memory.  "It's unlikely that analog EDA will ever reach the level of automation that's been achieved with digital design, so what we can expect are incremental improvements rather than a radical shift," he explains.

Still, if widely applied at various stages of the design process, incremental improvements could yield some pretty impressive "automation" results.  Gary Smith suggested the following Analog RTL workflow (elements that need significant development shown in boldface):

                               
Of these tasks, developing an acceptable level of analog synthesis is probably the most daunting task, according to Collins.  "EDA tools are only able to chip away at a design style that only gets more and more parametric and complex at each succeeding process node," he explains.  "To date every company that has tried an analog synthesis approach has failed [with] Barcelona being the latest victim."

Another reason analog synthesis may prove difficult is its dependence upon a usable gate level library. "Analog circuit design is generally non-linear, definitely non-Boolean, and is difficult and often impossible to constrain to a pre-designed set of cells in a library," explains Navraj Nandra, Synopsys's director of product marketing for Mixed-signal IP, "Analog design compilers [have been] fundamentally limited to circuits that performed only linear functions (i.e. the small signal behavior of op-amps) that described a very small set of problems in the field of analog design and thus not very valuable."

Still, there's hope some of these knotty problems can be solved over time, according to Filseth. "Interdigitating two transistors and making sure the structure conforms to both its nanometer design rules and its symmetry and other requirements, is largely done by hand today," he explains, "Yet the mechanics of moving polygons around and testing whether or not they pass DRC are well understood, so while it may or may not be possible to automate the entire physical design process end to end, certainly parts of it have the right characteristics."

EDA experts also believe that there's still plenty of "automation" mileage in the use of analog IP.  "Automation of mixed-signal designs has increased tremendously through the growth of mixed-signal interface IP," says Synopsys's Demler.  Similarly, Ashutosh Mauskar, vice president of product and business development at Magma Design Automation points out that reusable analog IP provides a basic level of automation.  "Designing any analog IP for the first time is a combination of art and science, but porting the analog IP from one process to another should not be," he says.

So, while it may take some time for automated analog design to reach anything close to the automation in the digital realm, there may be some (sub-wavelength) light at the end of the tunnel.  But until then, chip design firms (and EDA vendors) must continue to kowtow to analog designers determined that their craft should remain a "black art" for as long as possible.

ARTICLE SOURCES:

• Ashutosh Mauskar, Vice President of Product and Business Development, Magma Design Automation. Contact: Monica Marmie, monical@Magma-DA.COM
• Chris Collins, Director of Analog Technology Infrastructure, Texas Instruments, Inc.
• Eric Filseth, CEO of Ciranova
• Gary Smith, Gary Smith EDA, 408-985-2929, gary@garysmithEDA.com
• Jeff Lewis, VP of Marketing, Innovative Silicon
• Mike Demler, HSIMplus Product Marketing Manager, Synopsys. Contact: Sheryl Gulizia, Sheryl.Gulizia@synopsys.com
• Navraj Nandra, Director of Product Marketing for Mixed-signal IP, Synopsys. Contact: Sheryl Gulizia, Sheryl.Gulizia@synopsys.com
• Sergio Kusevitzky, Vice President of MIPS' Analog Business Group.
• Steve Lewis, Director of Product Marketing, Cadence