[content] => 
    [params] => Array
            [0] => /forum/index.php?threads/near-threshold-voltage-ntv-panel.13740/

    [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 (NTV) Panel

Daniel Nenni

Staff member
Where are the Opportunities – is it a “Free Lunch”? Part 1 of 3

Experts around the world helped to educate us during the 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 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. This is part one of a three-part series.

Brian: Some things are just too difficult until the way things were done in the past becomes even more difficult. That is certainly happening for many sectors of the semiconductor industry given the free ride of Moore’s Law is over for a large portion of the industry. When that happens, the difficult solutions become the more logical ones to try. Companies have to try harder to differentiate or they become commodities.

Performance has been the metric that has driven the industry for a long time but now power and energy are becoming bigger concerns for some markets. Where are the biggest opportunities for NTV and subthreshold design practices?

Energy has been a constant struggle for the mobile industry for many years, so it is not totally new. We also see the Internet of Things (IoT) evolving very quickly and these products will be less expensive than the maintenance of batteries. So, we have to make sure batteries last as long as the life of these IoT products or even there is no battery at all with energy harvested from the outside world. There is still opportunity to save power as everyone wants to consume less power. NTV is one of the ways possible to do really interesting things. There are plenty of applications where you want products to be on all the time, and it is probably in these areas where NTV is really interesting.

Lluis: We see a very strong demand for products running on a battery for years. So that requires we innovate – in processes, libraries, flows. The other area is the advanced processes, competing on lower voltage and DVFS. So, we see two areas, on the forefront where they are doing huge designs and on IoT, microcontroller-based applications.

Joachim: We have just heard about IoT and I think it is the biggest market. One very important market is biomedical devices - hearing aids/implants where we don’t need high performance. Another trend is embedded machine learning (ML) – sensors everywhere. Devices that will last for 10 and more years. Of course, Automotive is emerging with hundreds of sensors in the car where you do not want to drain the battery.

Lauri: All the applications have been mentioned but one of the things I have been seeing is the software part are disconnected from hardware. Software guys take as much of the power available and will be left wanting for more. This applies to all the applications - the software developers have to understand more about the capabilities of the hardware to take advantage of NTV.

Brian: There is never a free lunch – you gain something, you pay a price for it. Can we quantity it? What are the potential gains, what is the price of NTV design?

The first order is performance. People need some guidance so what we tell them is to look at leakage. If you can reduce leakage by 10X, probably performance will probably go down by 10X. If you can reduce leakage by 100x, performance will go down by 100X. An example on our 22nm, we can get memories close to 2GHz and we have memories for retention and low voltage. That same memory when optimized for retention and leakage only, we get 30MHz. That is a 100x right there. Of course, buying expensive wafers to run at 30MHz is not economically practical. So, you need to bring complexity in to recover some of that performance in different modes, where you go fast, where you go slow and from one to another. First order is Performance, second order is complexity – it’s a curve and you choose where you want to put your design.

Mike: NTV is interesting where you want to do processing in the background without having some large power consumption. In general, it’s a hierarchy where you have to wake up some portion of the circuit to do more processing when you detect something. It’s not a free lunch because you have to add constraints to your system to manage NTV.

Lauri: It’s not hard to get one chip working on NTV, but to do it and hit time-to-market and yield specifications – that’s where the difficulties and complexities come in.

Joachim: It depends on the application, NTV computing doesn’t fit everybody. Don’t try to get high performance with NTV computing. This is going to be very expensive so why do it if you’re going for high performance. But all these medical applications – hearing aids, pacemakers, implantables – are going to be a huge market.

Part 2
Part 3
Last edited: