August 17, 2018
Can you guess how much performance gain could be realized by using analog signal processing? How much lower power dissipation? Just how crazy do you think I am?
This is part eighteen of a series. Read part seventeen here.
Now that I am wrapping up my discussion on IoT, it feels that there is something I have missed; and that is a peek into the future as I see it. Much of the discussion over this blog series has had a theme of performance and power dissipation. The IoT system (or system of systems) needs to be optimum across the clouds, through the aggregators, and down to the smart sensors. In the near future, aggregators of today like the OSD335x System-in-Package, will become smarter, smaller, and lower power while enveloping much of the functions of the clouds and smart sensors.
The bigger hurdle of the IoT future will be in how we process all of this data we are collecting to get interesting information and take the appropriate actions. Further, we will need to do it efficiently, in real-time, while not throwing more energy consumption and performance at it.
For our roadmap to the future to make the necessary big leaps, we will need to make some fundamental changes.
Perhaps this is the time for me to introduce my thoughts on how performance will increase at the same time power dissipation will decrease. Allow me to do this from the perspective of my background: digital signal processing.
I spent the early part of my career at Texas Instruments telling university professors and industrial leaders that it was time to get rid of analog and fully embrace digital. We were introducing the digital signal processor device which we believed was the answer – now what was your question? The DSP, with its hardware multiplier, revolutionized the way interesting signals could be processed
I now believe that we are near the end of the era in which digital processors are the answer to the question (you get to ask the question, but the answer is always digital). We are at the beginning of a new era in which the answer will move toward being analog rather than digital. I can summarize this transition by telling a quick story:
Many years ago, I was asked to help my boss put together a presentation he was to make at MIT. I proposed a title of “TI made a ‘D’ in Signal Processing and is proud of it”. I think it is time for us to bring that grade up to an “A”.
Earlier this year, I was part of a panel discussion at the International Conference for Acoustics, Speech and Signal Processing (ICASSP) and presented analog computing as a new challenge to our industry and research community. I estimated the advantages of analog could give us 3 or 4 orders of magnitude higher performance, 3 or 4 orders of magnitude lower power dissipation, and 2 or more orders of magnitude in savings of size. My conclusion was “with the increase of performance is it worth finding solutions to the issues that need to be resolved?”
Allow me to present my challenge for IoT in a more generic way. Is it possible to obtain infinite performance while dissipating no energy in a dust sized package? And let me suggest we know how to do it already.
As I have said many times in the past “DSP is dead, long live DSP”, I now have to correct it to say “DSP is dead, long live Signal Processing”.
Here are a few final questions:
- Just how crazy do you think I am (limit your discussion to my thoughts on analog signal processing)?
- Where would be the first place you would consider using analog signal processing rather than digital signal processing? Why?
- Do you have a guess as to how much performance gain could be realized by going to analog signal processing?
- Do you have a guess as to how much lower power dissipation would be realized by going to analog signal processing?