EEVblog Electronics Community Forum
EEVblog => EEVblog Specific => Topic started by: EEVblog on March 31, 2021, 10:02:19 am
-
Halve Your Processor Power Consumption using Diffusion Capacitance!
Dave demonstrates a neat but tricky technique to PWM a processor's power pin to reduce power consumption.
https://en.wikipedia.org/wiki/Diffusion_capacitance (https://en.wikipedia.org/wiki/Diffusion_capacitance)
MIT: http://web.mit.edu/6.012/www/SP07-L16.pdf (http://web.mit.edu/6.012/www/SP07-L16.pdf)
Colorado University: http://ecee.colorado.edu/~bart/ecen3320/f99/pdf/ch4_4_5.pdf (http://ecee.colorado.edu/~bart/ecen3320/f99/pdf/ch4_4_5.pdf)
https://www.youtube.com/watch?v=PNWmfdH_Bzs (https://www.youtube.com/watch?v=PNWmfdH_Bzs)
Power a micro with no power pins! https://www.youtube.com/watch?v=2yFh7Vv0Paw (https://www.youtube.com/watch?v=2yFh7Vv0Paw)
-
I seriously do not buy it until I try it. Chips usually have some bulk capacitance inside on the power rails but that´s just ridiculous... And the switching frequency is quite low.
And you are driving a whole LCD bus let alone the processor itself. Edit: Which as stated below could be powering the processor.
And it´s April first coming fast to Aussieland so... no... Something is up... It can´t be that nobody ever tried that before, I have been on the internets for a while...
No no no no no... I saw quite well how good your acting was on the Keysight video... no no no no...
Edit: Actually reading a character from the LCD and then going to sleep plus the PWM BS, could also explain the low PWM frequency... really good call from the guy below!
-
It is not diffusion capacitance. Diffusion capacitance is due to minority carrier storage of forward biased junctions. You are using the the reverse biased substrate junction capacitors. They are never forward biased.
You are lowering the current since the PWM is lowering the average voltage the microcontroller sees. The internal voltage will have a sawtooth edge.
-
So in this example processor is actually takes it's power from SPI/whatever lines from display board if I understand the joke right? :)
-
So in this example processor is actually takes it's power from SPI/whatever lines from display board if I understand the joke right? :)
Good call! But that really depends on the LCD driver, you got to get the LCD in write mode for that to happen! Which of course it´s pretty reasonable!
-
I seriously do not buy it until I try it.
Look what timezone Dave is in and what date it is in Australia.. >:D
-
Assuming that the internal capacitance would really allow that and ignoring all the problems associated: how would that differ from simply lowering supply voltage? ATmega328P has pretty linear dependency of current consumed and supply voltage. My bullshit sense is tingling.
-
I seriously do not buy it until I try it.
Look what timezone Dave is in and what date it is in Australia.. >:D
I´ve seen many strange things in my life ok... enough to doubt what I know... and I know brown-out can be disabled in AVRs... But sure the Aussie is trying to pull a Roo on us... :-DD :-DD
-
Fascinating until I looked at the date.
Dave I think you do the best ones on you tube :-+
-
He was like 3 hours early... :)
-
This sounds just like capacitive deractance at work
-
Its not just microcontrollers it works on.
I've been using this technique on my core i7 desktop, PWM'ing the power at 11.574 uHz
-
To be honest, I liked MickMake's video better.
-
I used to subscribe, but Mick hasn't posted since a year ago, anyone know if he's ok?
-
I used to subscribe, but Mick hasn't posted since a year ago, anyone know if he's ok?
He's tweeting all right.
-
That was a fun one.
-
It can be done. Using the following device, you really can cut your electricity bills in half.
(https://www.eevblog.com/forum/blog/eevblog-1384-halve-your-processor-power-consumption!/?action=dlattach;attach=1206526;image)
-
I smelled something fishy as soon as I noticed his DaveCad of the CMOS Inverter looked a little bass ackwards.
Looked more like a buffer, eah?
Plus.. My calender was close by..
-
He was like 3 hours early... :)
No, I'm alway on time.
https://www.youtube.com/watch?v=iSHHe-AXxqw (https://www.youtube.com/watch?v=iSHHe-AXxqw)
-
Oh Dave, you get me every year. Although I worked this one out before the end of the video. Good work!
-
Its not just microcontrollers it works on.
I've been using this technique on my core i7 desktop, PWM'ing the power at 11.574 uHz
What duty cycle?
-
I have to admit: The Troll in the thumbnail was a dead giveaway. ;) But that might also be because i know those hideous things from when i was a child, they were all the rage back then.
But again i have to congratulate: You can get stuff like this across so professionally, that I can really believe them, being the layman that i am.
-
Its not just microcontrollers it works on.
I've been using this technique on my core i7 desktop, PWM'ing the power at 11.574 uHz
What duty cycle?
30-40% on time.
-
<insert the sound of slow clapping here>
I smelled something is wrong, but haven’t realized it’s April’s Fools. Dave got me. Good job. :D
-
He was like 3 hours early... :)
When I first saw the video I thought it was a joke as my calendar had already ticked over to April 1st. But I am in NZ and and we are 2 hours ahead of Australia so it would have still been 31st March in Australia when Dave released the video.
-
He was like 3 hours early... :)
No, I'm alway on time.
What time zone are you in in Sydney? Yesterday, Google said that 7am my time when I wrote that comment was 12am there, (today it is now 8:30am, April 1st here and Google says it is currently 1:30am in Sydney) and you has posted this topic at 4:02 my time, which was three hours prior.
Is Google's "Sydney Time" wrong or are you somewhere else?
-
Halve Your Processor Power Consumption using Diffusion Capacitance!
Dave demonstrates a neat but tricky technique to PWM a processor's power pin to reduce power consumption.
Yes, that or a similar technique is/was used in an early Clive Sinclair Calculator. It might be before your time and/or too British (i.e. It may not have reached Australia ?), for you to know about it.
http://www.vintagecalculators.com/html/sinclair_-_the_pocket_calculator.html (http://www.vintagecalculators.com/html/sinclair_-_the_pocket_calculator.html)
Clive Sinclair is an electronics genius and is noted for his miniature electronic products achieved by using very innovative designs. The first Sinclair Executive calculators used one of the standard calculator chips of the time, the Texas Instruments TMS1802. This normally has a current consumption of about 300 mA which dictates that normal AA size batteries, or larger, need to be used to give a decent battery life.
Sinclair in his quest for miniaturisation, to produce a true pocket calculator, wanted to use button cells rather than normal batteries, but these would be drained in minutes by the chip and the LED display. However, Chris Curry and Jim Westwood at Sinclair Radionics, in Huntingdon, England, found that the power to the chip did not have to be on continuously, it could be pulsed and the internal capacitance of the chip would store enough electrical charge to keep it working till the next power pulse. Power pulses lasting 1.7 microseconds are used, at a frequency of 200 KHz during calculations and 15KHz between each operation, reducing the power consumption to 25 to 30 mW. Texas Instruments did not recommend operating in this way, but it allows the Executive to get about 20 hours continuous operation from 3 small mercury button cells.
Another link:
https://www.i-programmer.info/history/people/370-clive-sinclair.html?start=1 (https://www.i-programmer.info/history/people/370-clive-sinclair.html?start=1)
It used a standard chip with a, typical Sinclair, power saving trick of pulsing the power supply.
N.B. The links are NOT April 1st jokes!
-
Since I watched this on the last day of March, I am glad it is revealed so that I do not have to explain what was really going on. The oscilloscope showing the gate drive waveform explained all, as I have seen that before.
-
Is the 328P actually running with just 1.25V (25% duty cycle of 5V) at 16 Mhz or this part of the fake?
I can’t tell if it’s fake or not, but the supply voltage wouldn’t be 1.25V.
This is not PWM smoothed out by a low-pass filter: there is no “pull down” phase. The supply cycles between high and HiZ, not high and low. That means any capacitance is instantly charged to Vcc during the on phase, but is later slowly discharged only by the current consumed by the controlled itself. The voltage is determined by power consumption of the controller and timing, not directly by the duty cycle.
-
The slowly falling supply voltage because of the "diffusion" capacitance is what Dave tells. The main part of the power is likely comming from the LCD interface, powering the µC with some 4 V or so.
The the voltage switches between some 5 V and some 4 V both a fine, though the fast transients may cause glitches. Similar powering via a supply pin can work but is still outside the specs and it can also cause funny behaviour, especiall with pins close by.
The odd point is not so much why the µC still runs with the PWM active, but why it fails with a lower PWM setting.
-
Pwm in Davecad is an N-channel fet. So DIO must be > 5V to turn on. But DIO is powered by the mcu itself, so I can't see how it ever switches on.
Do I read that wrong?
-
I'm somewhat disappointed, that the new turboencabulator wasn't used in this test setup.
-
I'm somewhat disappointed, that the new turboencabulator wasn't used in this test setup.
I am disappointed that Dave's old turboencabulator was not used.
-
I'm somewhat disappointed, that the new turboencabulator wasn't used in this test setup.
It was. It was secretly hidden under the bench, and was able to power the Arduino, via its flux-capacitor ultra-field rays.
-
Easy;
It is the hook up cable’s negative resistance. The load-side voltage becomes larger than the supply side.
Think of it as a wire with gain.
-
:blah: :bullshit: :blah:
I'll believe this load of bollocks when I see the power measurements done on a high-quality (eg, "Made in USA"), quantum-entangled, reverse neutron polarity, sub-fractional cardinal-grammeter. :-DMM
-
that's why we (the semiconductor farmers) have been using fully depleted silicon substrates for at least he last decade. SOI (Silicon on insulator) does away with Csb and Cdb so the floating node can switch much faster.
You can also make structures denser ( not smaller. this i has to do with the distances between diffusions , not the size of the diffusion . size is current. distance is voltage. )
All those processors running your smartphones are built using fully depleted silicon.
of course your average arduino is not built using that technology. that barely left the vacuum tube age.
http://csit-sun.pub.ro/courses/vlsi/Carte_VLSI/bookch15.pdf (http://csit-sun.pub.ro/courses/vlsi/Carte_VLSI/bookch15.pdf)
-
of course your average arduino is not built using that technology. that barely left the vacuum tube age.
I suspect SOI is limited to higher margin products which can afford it. Microcontrollers also require floating gate memory which is only available on processes with a much larger feature size; I think they have only just reached 28 nanometers.
-
He was like 3 hours early... :)
No, I'm alway on time.
What time zone are you in in Sydney? Yesterday, Google said that 7am my time when I wrote that comment was 12am there, (today it is now 8:30am, April 1st here and Google says it is currently 1:30am in Sydney) and you has posted this topic at 4:02 my time, which was three hours prior.
Is Google's "Sydney Time" wrong or are you somewhere else?
I care about my audience everywhere around the world.
https://www.youtube.com/watch?v=iSHHe-AXxqw (https://www.youtube.com/watch?v=iSHHe-AXxqw)
-
Pwm in Davecad is an N-channel fet. So DIO must be > 5V to turn on. But DIO is powered by the mcu itself, so I can't see how it ever switches on.
Do I read that wrong?
Nope, I goofed.
I realised this before I shot and drew it again, then dumb arse me used the old sheet in the video instead of the new correct one :palm:
-
Ahh, the intent was clear, but since this was posted as a (semi) riddle I wasn't sure that it might not be a clue.
I wonder, given stray cable inductance/ or from a choke somewhere + blocking diode + fet, it resembles a simple switching converter.
And extra capacitance from the LCD board decoupling would help to store energy.
-
Ahh, the intent was clear, but since this was posted as a (semi) riddle I wasn't sure that it might not be a clue.
I did notice it during editing but then realised it would be fun to leave it in rather than reshoot, so ;D
-
Do you have to disable brownout detection to get this to work?
-
Interesting phenomenon. Certainly not recommended for anything serious.
Most MPU's at least have plenty of different power modes which should be managed as required.
I worked in the electronics team designing Zoleo (www.zoleo.com (http://www.zoleo.com)). The device talks to the Iridium satellite network, GPS satellites, a mobile phone, and a host of internal I/O devices. From a small single cell 3.7V battery this baby will typically operate for an impressive 200 hours between charges. One task I had was the power consumption calculations during the design, and it was not easy by any means. Every chip, pullup/pulldown resistor was analysed prior to the schematic being locked down. And within the MPU, the power consumption with different power modes under firmware control was critical to keeping current drain minimal. With most products no one cares much about a few milliamps here or there, but this application was critical because as Zoleo is used by trekkers, often in the middle of nowhere. This product is now a world beater which is a credit to all on the R & D team. A very satisfying result and a terrific product even if I say so myself.
Most small CPU's these days consume very little power and can often be left in sleep mode until an external interrupt, from, say a MEMS device springs it into action. Using PWM to reduce power consumption on a micro in this day and age for a commercial product is nonsense. Many products do not have anywhere the level of power consumption analysis done that we did for Zoleo. The importance of this of course depends largely on the application requirements.