[content] => 
    [params] => Array
            [0] => /forum/index.php?threads/near-threshold-voltage-a-much-needed-reality-or-risky-dream.13741/

    [addOns] => Array
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2020570
            [XFI] => 1050070

    [wordpress] => /var/www/html

Near Threshold Voltage: A Much Needed Reality or Risky Dream?

Daniel Nenni

Staff member
Near-Threshold Voltage (NTV) Design Panel: Are designers ready – are the tools ready?

Experts around the world helped to educate us during the 2020 Virtual 57th Design Automation Conference (DAC) on where NTV is headed, why it’s it not a free lunch and how the industry is working together to make it work for certain applications. This is part 2 of a 3-part summary of the lively video discussion that took place between Brian Bailey, Semiconductor Engineering; Lauri Koskinen, Minima Processor; Mike Eftimakis, Arm; Joachim Rodrigues, Xenergic; and Lluis Paris, TSMC.

Brian: Lauri, you’ve had some experience in both academic and industrial environments. How ready are designers for this? What new skills are they going to need to learn to do this kind of design?

I’d say that in the cutting-edge companies like big mobile or big processor companies, you’ll find designers who are very ready for this. If you go into the basic embedded companies who are good at time-to-market (TTM), then signing off for different voltages, more complex constraining in place & route, and not over margining everything, will be deficiencies in the design skills The challenge is getting that leading-edge knowledge into the smaller companies who are looking at TTM, so that it doesn’t become a risk in their next project or design.

Joachim: I totally agree with Lauri – companies would like a push button approach. No manual interference, it has to be seamless. It should not be disruptive at all. This is what companies are asking for. That’s about quality, it’s TTM.

LLuis: That sounds pretty much like a free lunch! There are compromises. What we have to do is have solutions that work easily for some applications and extend from there. And the whole industry and corporate learning is going to help. Look at applications where it’s easier to do and grow from there.

Mike: It’s also a question of adapting your application and understanding how you can make it work. We have demonstrated recently we can run keyword spotting at 2MHz with one classification per second. It’s degraded performance compared to what usually you would do, but you can achieve interesting results and you can run that at NTV. You can get interesting results but that means you have to review the whole flow from the process technology to the software and understand what needs to be adapted to make it work.

Brian: What about the tools for NTV?

The tools are improving. They are improving a lot. Synthesis with the Liberty Variation Format (LVF) flow is pretty much there. What still needs help is library characterization tools. That’s the next frontier. We’re telling customers to use this version of the tool, use this configuration. TSMC is providing packages on exactly how to configure the libraries. Designers who have done this before know how to do it but new customers don’t. We’re bringing that methodology together with tools. We think that’s the way to go together with libraries characterized for low voltage. We think the three elements – tools, libraries characterized, methodologies – that should open the door.

Lauri: I agree with Lluis – he knows the pain points. There have been very small improvements in the library characterization tools. They are a big pain point still. You change the version of the tool and you get a different result. Definitely, that’s the first place that needs improving.

Joachim: It takes a lot to get a proper flow up and running. You need to know what you’re doing. With TTM, companies are reluctant to just give it a try. As soon as there’s a larger market, the EDA companies will see the potential of this technology. The breakthrough could come as we’re working together, the IP providers, the technology providers. We’re using gates not optimized for NTV. If we could get more NTV-optimized libraries from the beginning, we could have better possibilities.

Lluis: We’re working on that. We think we’re doing that. We’re creating libraries, we’re preparing a methodology package and giving to partners. Many partners are doing libraries. We are explaining what we are doing with customers, customers seem to be happy with that. Then we’re applying that to the industry. We’re in the wafer business, not the library business. What we are learning, we are teaching. What we know, we are sharing. You can always do more, but we think things are getting better.

Part 1
Part 3
Last edited: