The IEEE Standards organization has unwillingly been pulled into the tug-of-war between the Common Power Format (CPF) and the Unified Power Format (UPF). The former, you probably remember, is sponsored by Cadence under the auspices of Si2, while the latter, developed in an Accellera working group is favored by the other "big three" EDA companies.
A short background is necessary. Cadence internally developed a format to use with its EDA tools to specify low-power requirements and constraints. At DAC last year it invited both EDA tools providers and users to join it in an organization called the Power Forward Initiative. After some discussion Magma, Mentor, and Synopsys declined to join the initiative because they had no access to the specification without signing a non-disclosure agreement and because Cadence wanted to maintain editorial control of the specification. Cadence then tried to have its specification adopted as an IEEE standard but ran afoul of IEEE procedural rules and abandoned the effort. It found a willing partner in its effort to "standardize" its document in Si2, with whom it has worked successfully with its Open Access initiative. Si2 formed a group called the Low Power Coalition and this group slightly modified the Cadence specification to arrive at a public document that Si2 called the Common Power Format. It is interesting to note that PFI continues to exists in parallel with LPC. Only eight of the twenty PFI members are part of the LPC membership now amounting to 17 companies. LPC does not represent a majority of the PFI group. If LPC were a true step toward fulfilling the goals of PFI, I would think all of the PFI members would have wanted to join LPC.
CPF is an Si2 specification, not a standard since Si2 is not an officially recognized standard making organization. In parallel, Accellera formed its own working group and developed its own specification: UPF. Accellera is also not an officially recognized standard making organization but, contrary to Si2, it has a long-standing track record as a "feeder" organization to the IEEE. This means that Accellera almost always will offer its specifications for standardization with no strings attached, as required by IEEE rules. Once Accellera approved UPF it offered it to the IEEE through the Design Automation Standard Committee (DASC). At the moment neither Cadence nor Si2 are willing to offer CPF to the IEEE.
Eric Filseth, corporate vice president at Cadence, in answer to a written set of questions from me wrote: " As a technical achievement, we're proud of it -- it works and supports real chip designs today. As an industry format, it's been surprisingly controversial, to us as much as anybody; we're software engineers, not experts on IEEE rules." I agree that Cadence should be proud of its technical achievements and should want it transformed into an international standard. What is surprising to me is the last part of his answer. Victor Berman, chair of the DASC, was an employee of Cadence until early this year, and Stan Krolikoski who has joined Cadence after Victor's departure as Group Director,
Standards and Interoperability, has been involved in IEEE standard development within the DASC for most of his career, starting before 1987. Either gentlemen could have given an authoritative tutorial to any one in Cadence about IEEE rules and procedures. I find it also hard to believe that Cadence as an active member of the IEEE Corporate Advisory Group (CAG) does not know its standardization procedures!
A matter of "convergence"
At the last meeting of the IEEE New Standards Committee (NesCom), its members received a copy of an email from Cadence stating that "We at Cadence have extreme reservations about the proposed 1801 low power PAR. We believe that approval of this PAR at this time will be detrimental to both the Electronics and the Electronic Design Automation industries."
Although both Cadence and Si2 have, at different times, cited different reasons for their decision, when asked to justify their concerns Cadence agreed with Si2 that the Si2 convergence methods are superior to those of the IEEE. Si2 maintains that it admires the IEEE and wants to co-exist peacefully with it, since "many of our members are also members of the IEEE" according to Steve Schulz, president and CEO of Si2. What struck me as odd in this statement is that Si2 itself is a member of the IEEE, a fact that neither Cadence nor Si2 has ever pointed out publicly to my knowledge. Convergence is important because national and international standards are founded on the principle of consensus. Although not all voting members need to vote for a standard, in most cases 75% majority is required, and all technical issues must be resolved before a proposal is accepted as a standard. The IEEE has two voting methods: either individual or company. In the first case all qualified professionals have the right to vote as individuals. In the latter, each company casts one vote. Company voting was introduced to the IEEE with the formation of the CAG. Both Cadence and Si2 are members of the CAG, as is Accellera.
Yet, Steve Schulz is not the first one to blame the IEEE for the slow progress some of its standard development working groups make. Aart de Geus has said it, and so did many at Cadence in the matter of Verilog. Achieving consensus, especially in areas that are either technically or commercially controversial is not easy. As a matter of fact there is no consensus on low-power format, in spite of Si2 efforts to lead the work. When asked to define convergence, the necessary step to consensus, as it applies to Si2 Steve Schulz provided me with a written answer, reported below in its entirety.
" Convergence at Si2 takes a proactive approach to engage with key influencers across the supply chain, who are best equipped to replace artificial barriers to alignment with more genuine and significant business incentives to come together for a common result. Si2 staff have organizational incentives to produce results.
The approach is to: (a) align with market-leading end-users; (b) proactively seek to partner with all relevant suppliers; (c) focus beyond near-term divisive issues toward longer-term shared objectives (i.e. working flows).
I will note that, at the formation of the LPC, we received signed letters of intent from all of the Big Four EDA vendors that they would join the LPC and help set its direction. However, this did not occur and things became increasingly polarized over time. The industry was unable to bring the groups together at this point. There will possibly be more opportunities ahead."
Some people in the industry believe that I am exceedingly "anti-Si2". But I am only observing and reporting facts. Take the statements above. Si2 strives to: "engage with key influencers across the supply chain, who are best equipped to replace artificial barriers to alignment with more genuine and significant business incentives to come together for a common result." And: " I will note that, at the formation of the LPC, we received signed letters of intent from all of the Big Four EDA vendors that they would join the LPC and help set its direction. However, this did not occur and things became increasingly polarized over time." Taken together these statements show a lack of leadership and ability to reach consensus, the foundation of a working international standard. The principal reason for the failure was Cadence refusal to admit the other three, or anybody else for that matter, as equal partners in the development of the specification. The document released by Si2 after the Cadence announcement that it had implemented the format, matched that implementation exactly: a rare occasion if other technologies had been even considered in the development of the specification.
Another part of Steve's statement is surprisingly off target. He said: " Convergence at Si2 takes a proactive approach to engage with key influencers across the supply chain." Cadence's announced recently that TSMC had released the first implementation of the CPF specification in one of its 65-nm libraries. Yet TSMC is not a member of the LPC, although it is a member of PFI. Are we to deduct that TSMC is not a key influencer across the supply chain? Further, Synopsys has announced twenty successful tapeouts using its own low-power technology now part of the UPF standard. Are we to assume that Synopsys is not a key influencer across the supply chain?
Unfortunately, the longer Cadence and Si2 work to justify an illogical position with respect to the low power format specification, the more in trouble they get with semantics.
What is peculiar here is that the issue is now one of "personalities" not of technology or even standardization process. Both Cadence and Si2 want to be recognized as important contributors to EDA. They are: so let's stop feeling inferior and let's start working together. Cadence made Verilog what it is today, and Si2 is a good steward of the Open Access effort: those alone should be enough to restore some self-esteem. I believe Cadence's format is technically strong and can hold its own in the midst of enlighten technologists. Here is my counsel: have no fear -- take it to the IEEE and prove it.