This is part sixteen of a series. Read part fifteen here.
We had been shipping our first DSP, the TMS32010, for several years and were introducing the second generation along with CMOS versions of the first generation. At the time, twenty-five years ago, all of our devices were designed to work at five volts, as was the standard for the whole SC industry. However, later we found out we had an opportunity (or as initially thought, a problem) when a customer using our devices to design hearing aids gave us a call. They wanted to buy "more of the three-volt version" of our DSP. We were polite but assured them that we only made five-volt versions. But they were more than certain we had a three-volt version. It seems that they had purchased quite a few of our devices and all had worked at three volts, so they were certain we made them. We hung-up the phone and walked away asking ourselves what we should do next: advise the customer to stop using the device out of spec, or evaluate the new business opportunity of low-power?
My first action was to assign one of our application engineers to characterize the specific device the customer wanted to buy. The resulting characterization agreed with the customer and so generated my second action. That second action was to create a data sheet for a three-volt version for the device. Once we were all convinced we could actually manufacture it and got over the minor issue that none of the complementary devices we sold (e.g., TTL, Memories, etc.) could interface with a three volt part, we reconnected with the customer and began shipping them devices. I don’t remember the price we put on them, but I suspect it was at a higher price than our five-volt version.
This was the beginning of the journey I took inside of the company and in the industry to make power dissipation a performance metric of the DSP. We had discussed the three volt opportunity with our IC designers and our silicon process engineers. They all agreed it was possible to move to three volts with no issues, but by that time I was already considering the idea of how to go to two volts. When I asked about going to two volts, I got the expected answer that the process parameters would need to be changed a bit to give a yielding two volt part but very possible. But I was well ahead of them and thinking about one volt operation. Again, they gave the tactical solution for a one volt part. I was beginning to think I wasn’t casting the vision well enough for them as my goal was to continue down to lower and lower voltages. My last request of them was to get their input on how we could create devices that ran on “body heat”. I believe they were catching my vision while privately labeling me as crazy.
Later, I took the whole idea to the rest of the industry with a chart I created and called Gene’s Law. Figure 1 shows the chart I used for Gene’s Law, showing the idea that power dissipation would be the next performance metric in the world of DSP (and beyond DSP to the whole industry). Simply stated the chart shows the relationship between the amount of power required to perform the basic multiply and add function in a processing element to time. The data suggests that the unit of power dissipation to perform the basic element of a multiply/add (MAC) was halved every 18 months.
[Figure 1 | Gene’s Law, taken from a presentation at ISLPED in 2005.]
I’ve told this story to get us to a point at which we can talk about the power dissipation goal of a smart sensor. And, yes, the answer is body heat, or a practical version of body heat suggested by the idea of Smart Dust: as much energy as can be scavenged from the environment.
For completeness, I must recognize some work done at MIT several years ago to develop an "Ultra Low Power Harvester Power Management IC with Boost Charger, and Autonomous Power Multiplexor" device, later called BQ25505 under Dr. Anantha Chandrakasan. Below is a picture taken at ISSCC 2012 when the work was presented (Paper 5.8). Figure 2 shows the student, Karthik, describing the device to us.
[Figure 2 | ISSCC 2012 introduction of a power management device. From left to right: Fernando Mujica, Brian Chan, Karthik Kadirvel, Gene Frantz, Yogesh Ramadass.]
Here are some questions for you to think about:
- What is your definition of “body heat”? (Please do not refer to the movie by the same name as your reference point.)
- If a battery is the source of the power, how long should the battery life be to be considered a practical equivalent of body heat? Look back at a previous blog in this series (15) for a hint to how I would answer it.