In my thesis I compare conventional weighed binary processing with time-encoded stochastic computing. This includes power consumption during processing as well as power consumption in the analog-to-digital conversion. I use two LTC6992s to generate voltage controlled PWM signals (192kHz and 180kHz) and one ADC (LTC6992) for comparison. I would have loved to find out that I can save power by replacing conventional ADC’s with voltage-controlled pulse-with-generators.
However, the data sheets say (and my measurements confirm) that the LTC6992 requires more power than the LTC1098 (@5V). Approximately 0.5mW for the LTC1098 and 1.75mW for each LTC6992. Now I am searching for possible explanations.
Thank you in advance.
The LTC1098 very likely uses an older process node, but I don't think we really provide that information - what counts in the end is the performance (including power consumption) that we can guarantee in…
Are you trying to create an ADC? For power consumption, it will be difficult to do better than a modern SAR ADC.
The LTC6992 is a wonderful part, but it is indeed a bit of a "Swiss Army Knife", not optimized for ultra low power consumption.
Perhaps you could use the LTC6992 for your proof of concept, then move to an optimized solution later. A quick web search will turn up lots of integrated circuits as well as discrete designs, you'll have to decide which meets your design constraints.
Hi Mark, thank you for replying.
I am not trying to create an ADC. I want to do power efficient digital computations without them. I do research in stochastic computing, which is a promising alternative computing paradigm that works with streams of bits (bit-streams) instead of traditional weighed binary numbers.
Usually, researchers use conventional ADCs combined with digital circuits inside FPGA’s or ASICS to generate the bit-streams. However, this approach sometimes consumes more power than we can save by doing approximate computing with stochastic bit-streams.
I already used the LTC6992 as proof of concept to show that in stochastic computing we can replace the ADC (LTC1098) plus digital circuit by a single PWM module. I think the question “Does PWM generation require more power than AD conversion” would be a whole thesis on its own because it depends on so many factors (latency, frequency, resolution, process technology, etc). But If you say it is difficult to do better than modern ADCs, I can consider another point of view.
I was hoping to get information like “The LTC6992 uses xxx-nm and the LTC1098 uses a newer/older technology”. A found a press release of the LTC6992 from 2010 and an old datasheet of LTC1098 (before 2000). I am not sure if this says that the LTC1098 uses an older technology.
The LTC1098 very likely uses an older process node, but I don't think we really provide that information - what counts in the end is the performance (including power consumption) that we can guarantee in the datasheet. There is a figure of merit for ADCs called the "Walden" FOM that you might be able to calculate for candidate ADCs, then try to do the same thing for the PWM generation circuit to come up with some direct comparison. But as you mention, that's a thesis on its own and out of scope for this forum.
On the PWM approach, again, you can probably do a lot better than the LTC6992 with a discrete PWM generator circuit, and there are lots of circuits out there you can find with a quick web search.