1

I am learning about ultra low power (ULP) microcontrollers and I've listed some criteria:

  • Current consumption with only RAM retention (~70nA)
  • Current consumption with RAM retention + RTC (~150nA)
  • Minimum voltage (~1.8V)

I don't often quite see the run mode consumption in μA/MHz. Furthermore, I assume for the same lithography process (40nm?), an 8-bit MCU would consume less than a 32-bit. Also an 8-bit is not always 4x faster than a 32-bit one, especially for IoT embedded applications.

Why are ULP microcontrollers usually only defined with the above three criteria?

JRE
  • 67,678
  • 8
  • 104
  • 179
nowox
  • 477
  • 4
  • 12
  • A 32-bit MCU can perform more operations in less cycles than a 8-bit MCU so I guess it depends. If your algorithm is mainly boolean (if condition then perform that, else perform this), an 8-bit CPU would probably be better. But if there are calculations involved, a 32-bit MCU could be better. – Ben Aug 24 '21 at 15:24
  • Is it true that a 32-bit MCU can perform more operations in fewer cycles? One ALU per core means one operation per cycle, isn't it? – nowox Aug 24 '21 at 15:27
  • 3
    It depends on how wide your data is. If you need to work on 32-bit data, the 8-bit ALU will require many more cycles than the 32-bit ALU. – Dave Tweed Aug 24 '21 at 15:46
  • 1
    Yes I agree but often in low power embedded devices you don’t need 32 bit operations – nowox Aug 24 '21 at 15:52
  • 2
    @nowox MSP430, a 16-bit MCU, does list itself in \$\mu\text{A}\$ per MHz, for example. And they consider themselves (sold-as) an ultra low power device. In fact, it was designed originally in Germany for exactly that reason. How modern tech gets applied will depend on where business sees its primary markets, though, and not necessarily where you may want it. – jonk Aug 24 '21 at 16:02
  • 1
    I guess a 16-bit MCU consumes even less, and 8-bit yet less - but it depends on the latest technology being used. There is a little market for 8-bit, so don't expect they gonna make a new series of them with the latest technology, unless there will be a such demand. The 16-bit are still being updated, can't beat the MSP430 I guess, it has all newest features like FRAM,... Here you go, newest products : https://eu.mouser.com/new/texas-instruments/ti-msp430fr-fram-mcus/ – Marko Buršič Aug 24 '21 at 17:20
  • @MarkoBuršič The MSP430FR is 1.5uA with RTC but the SiM3L1 is only 180nA with RTC. However, the MSP is better in the active mode (81.4 uA/MHz vs 114 uA/MHz) – nowox Aug 24 '21 at 20:30
  • Of course, there is a hard battle going on and hard to define which one is better: https://www.rs-online.com/designspark/which-low-power-mcu-is-really-the-lowest-power There are also commercial papers that tend to spit the competitors: https://www.ti.com/lit/pdf/slay015 – Marko Buršič Aug 24 '21 at 21:38

1 Answers1

3

Low power rating is a combination of the following (and more that I dont know about!)

  1. What fraction of the chip is actually used by the test. i.e, roughly instruction groups available Vs used by the test.
  2. How quickly the quantum of work required by the test can be done with the instruction set and data width. The test probably uses some 32 bit (guess)
  3. Sleep/standby mode efficiency
  4. The "leakiness" of the silicon during operation. Influenced by the process. Eg: 10nm chip will waste a lot less than 100nm silicon process chip in capacitance based energy loss.

At the moment, ARM 32 bit cores win in 1,2 and probably 4. So, until we get an 8 bit micro with the right instruction set, built on a small nm process, this will continue. Also, when the tests begin to take into account other peripherals on the chip, the numbers may change.

kalyanswaroop
  • 331
  • 1
  • 5