A Tale Of Two Standards

By Ed Sperling
It could well be one of the strangest developments in standards history. Two competing standards for power formats were rolled out in the middle of the last decade and aside from a few cries of foul they fell below the radar screen of most chip designers and architects for a half-dozen years.

Fast forward to the present and the Common Power Format (CPF) and Unified Power Format (UPF) have chipmakers up in arms, railing about competing standards and more than one version of even the same standard. And while significant work is under way to achieve interoperability across all of them, there isn’t even a hint that they could be merged into a single standard.

Why this sudden interest by chipmakers? It’s because mainstream has shifted from a 90nm process to 50/45/40nm. At the advanced geometries power needs to be considered up front in a design for the first time rather than addressed later in the design flow. Various modes of operation have to be considered, power islands need to be turned on and off, and issues such as electromagnetic interference, electromigration and electrostatic discharge are suddenly mainstream considerations. In short, you need to define power intent.

Understanding those issues at the architectural stage can save enormous amounts of pain—and money—further down in the flow, particularly at the verification stage and later at tapeout.

What’s good?
All of this isn’t necessarily bad, however. There are things to like about both standards, even if you don’t like the existence of two.

“There were a few of us complaining from the start that there should be only one standard,” said Vic Kulkarni, general manager and senior vice president of the RTL business unit at Apache Design Solutions. “As an independent vendor we have to support all the customers because both standards are getting adopted for low-power intent. But there are differences. CPF tends to be more back-end friendly and UPF is more front-end friendly.”

He noted that UPF, which has evolved into the IEEE 1801 standard as version 2.0, is the better format for verification. It provides options for supply network semantics, power switch modeling and simulation support in power states. It also provides extensions for logic designs with power-specific capabilities and constraints without modifying the original logic, and ways to stitch the supply port to the HDL port through concept of value conversion tables.

CPF, meanwhile, provides better support for library cell-related commands so it can define always-on cells, global cells and pad cells. Unlike 1801, the same CPF file can be used at all levels of the flow. And there are more back-end-specific commands involving IR drop limits and power nets.

Source: Apache

“The best compromise will be to build bridges between them,” Kulkarni said. “That’s been the focus of Si2, which has been working on bridges for power voltage analysis and voltage scaling.”

But there’s another school of thought that says competition in this market is a good thing. “The presence of competition may have made standards happen in the first place,” said Dennis Brophy, director of strategic business development at Mentor Graphics. “Consumers say they want one standard and that the standards we have are not pro-consumer. But this also has aggregated users to ensure the information they get is portable and useful and to make sure there is some way we can understand the semantics of one and import that into the syntax of the other.”

What’s bad?
Getting that type of automatic translation from one standard to the other isn’t easy, though. Qi Wang, technical marketing group marketing director for Cadence Solutions Marketing, likens it to providing an easy translation between Chinese and English. The syntax and semantics of each language are different.

“UPF 1.0 describes everything through power and ground nets,” said Wang. “CPF is more power-domain specific.”

Source: Cadence

That’s fine for companies that develop tools with either of those formats, but most large companies and many small companies have a combination of tools from multiple vendors. That makes things confusing enough, but it gets worse on the UPF side because there are two versions of UPF—version 1.0 and the more encompassing 1801—in the new 1801 standard.

“There’s a lot of legacy in 1801,” said Wang. “To really get to convergence you need to clean up 1801. The number of command options has increased five times.”

Synopsys sees things differently, of course. Synopsys and Mentor support 1801, while Cadence supports CPF. And while all support interoperability and overlap between the standards, there will still be two standards for the foreseeable future.

“There are issues being dealt with inside the standards committees, but part of this has to happen outside of what the standards committees can do,” said Yatin Trivedi,
director of standards and interoperability programs at Synopsys. “Synopsys supports a subset of what 1801 customers are asking for and Mentor may ask for a different subset. The overlap is probably 80% to 85%. Many Synopsys users are focused on certain types of design and power while at Mentor they may be focused on multivoltage designs.”

What becomes particularly problematic is third-party IP, although work is under way to eliminate that discrepancy. Synopsys is one of the largest IP vendors in the industry, particularly after its acquisition of Virage Logic, but the IP it acquired with Virage supports CPF while Synopsys is firmly behind the competing 1801.

“Moving forward, we see the format will not be the end,” Trivedi said. “It will only be the enabler. The first part will be understanding the capabilities of the formats and interoperability. The second part will be assessing whether users are feeling confident enough with methodology embedded into the tools.”

The future
Standards group Si2 has been working on bridging these two worlds in multiple areas. Steve Schulz, president and CEO of Si2, said the results of that work will be unveiled early next year.

Schulz noted that the starting point for interoperability is the 1801 standard. While these two standards will never completely converge there will be enough interoperability and an interoperability guide—as well as an understanding going forward that each standard needs to proceed with an awareness of the existence of the other. “This ought to get better over time,” Schulz said.

There are a lot of large chipmakers banking on that.