Before January comes to an end it is time for my annual flashback and brief reflection on where we are in system-level design, and a look at how the state of today compares to the predictions we made 10 years ago.
2011 was an interesting year for system-level design. In May Cadence announced its participation in the system-level domain with the System Development Suite. The year before, in 2010, we had seen consolidation in the virtual platform space with Synopsys picking up VaST, CoWare and Synfora. Later in 2011 we saw Mentor spinning out their high-level synthesis technology to Calypto. Overall, system-level design is a topic covered by all major vendors and not a forte of visionary start-ups anymore.
This year, when digging up the January IEEE Spectrum collection from my garage, I found issues from both 10 years and 15 years ago, 2002 and 1997. Ouch. 1997 was the year when I moved to the U.S. leaving active software and chip development behind, joining Cadence in technical marketing ... I have been in system-level design for 15 years. Time for a check point.
The 2002 issue of IEEE Spectrum was published in the wake of 9/11. It was understandably about smarter intelligence and improving security while protecting privacy, but also about global climate change, saving water supplies, safer car design and telemedicine for the masses. Sometime between 1997 and 2002 IEEE Spectrum had stopped talking about specific hardware and software design challenges in their January issues and instead moved to application-specific vertical views of the challenges ahead. My favorite article "Building Safer Cars" authored by Willie D. Jones talked about some of the design philosophies behind partitioning tasks between driver and car, and most of the safety mechanisms discussed then, like collision warnings, tile pressure sensors, and intelligent brake control are today either implemented or shown in advertisements for cars I may buy in the future. Bottom line -- it took 10 years for those features to become mainstream.
Opening the 1997 issue,was somehow déjà-vu all over again like in my previous reviews -- the topics of higher levels of abstraction were topics of discussion at least from 1997 to 1999. In his 1997 IEEE Spectrum article "SoftwareTools," Cary Trlica talks about "full systems designed with virtual prototypes." In a comparison of the major vendors at the time, he said that "Cadence is pushing the reusability of intellectual property in its block-based approach to system design." Note that this is prior to the VSI Alliance soon created after. Pointing to system design issues, the Cadence approach is characterized as "The designer develops functional blocks, builds the system, evaluates the interaction between the specific blocks and hence the overall system functionality." This is clearly a precursor of the upcoming Felix initiative and its resulting product VCC. Very forward looking, the approach is characterized as user being able to "leverage pricing and intellectual property for each piece and can rank them for cost and power."
Cary Trlica's 1997 article also states that "Behavioral design is the forte of Synopsys. The logic synthesis company stresses raising productivity by moving up from register-transfer level (RTL) design to do behavioral design, thereby greatly speeding creation, verification, and modifications of designs using high-level specifications."
Rolling forward 15 years, the issues mentioned are still big issues to be addressed. Today several vendors have high-level synthesis technology, which is used at the block level. Block re-use and integration are accepted issues, with both companies offering IP portfolios. Integration of blocks is still done mostly manually at the RTL level, although there is some automation provided here by companies like Duolog and Magillem. Bottom line, the issues foreseen 15 years ago were the correct issues. I would argue that complexity of design and the addition of software into the problem mix have prevented fully automated top to bottom flows.
So what has happened to the system-level design market since those predictions? In order to understand the dynamics we are talking about when "moving up", I took the 1994 to 2010 data from the EDAC consortium and to smooth the growth curves I calculated the 3 year CAGRs for the sub markets. The result can be seen in the graph associated with this post:
EDAC splits the overall EDA market into the following markets:
The resulting graph illustrates a couple of points:
Overall a lot has happened over the last 15 years, and some of the areas of ESL, like high-level synthesis have entered mainstream for the sub-markets they serve. The area remains exciting, especially given the ever more growing importance of software.
It will be a fun next 15 years ...