3 ms·

"Is there even a RISC-V that can compete against the ARM Cortex-M33? Or ARM Cortex-M0+ ?" Yes, of course, lots of them. That is where RISC-V has taken the most

"Is there even a RISC-V that can compete against the ARM Cortex-M33? Or ARM Cortex-M0+ ?" Yes, of course, lots of them. That is where RISC-V has taken the most market share from ARM so far. SiFive's E20, E21, E24 hit various points of that. Plus they offer the S21 which is a 64 bit microcontroller in Cortex-M33 class. ARM doesn't offer anything like that -- if you want 64 bits from ARM then you have to take the full no-subsets ARMv8-A ISA. The Europeans have their Ibex (formerly Zero-RISCy). Andes has the N22.

dragontamer 4y ago
What's the ULPMark (power benchmark) on any of those SiFive RISC-V chips? STM32U5 (one of the many ARM Cortex-M33 designs) is 464 ULPMark-CP. I don't think there's even a power benchmark on any of those chips you've mentioned. At least, I'm not seeing any online in my searches. STM32U5 (and most other Cortex-M33 based microcontrollers) are ~110nA of power consumption sleep mode, and 16uA / MHz at 160MHz, some of the most power-competitive chips I'm aware of on the market. I'm talking "run a microcontroller on a CR2032 cell for 10 years" kinda thing here. -------- EDIT: Looking at the ULPMark website (https://www.eembc.org/ulpmark/ulp-cp/scores.php https://www.eembc.org/ulpmark/ulp-cp/scores.php), it looks like the #1 chip is RSL15, which is ARM Cortex-M33, as I expected. With a ULPMark score of 1090 (higher the score, the less power the device uses). Nearly everything on the top list is Cortex-M33... Except LP5100 and XRM32UL051 are some Chinese Chips I don't understand. They got some impressive scores but I don't know what architecture they have. Nearly everything else is Cortex-M33 (except for the Ambiq, which is ARM Cortex-M4). Cortex-M0+ was the king of this a few years ago, but M33 has displaced it. We still see Cortex-M0+ chips due to much lower costs around though. So if RISC-V is around Cortex-M0+ level, I'd still be interested.
- brucehoult 4y ago
  "What's the ULPMark (power benchmark) on any of those SiFive RISC-V chips?" SiFive, like ARM, doesn't make chips, they license cores to people who make chips. The power characteristics are up to the chip designer -- what node they use, which "corner" of that node they use, and various other chip design features -- and not the core designer. I don't know of anyone who is putting out stand-alone microcontroller chips using SiFive cores. They've been concentrating on getting them into larger systems. For example Qualcomm said a few months ago they've shipped 650 million SoCs (i.e. Snapdragon XXX) with RISC-V cores in them (probably SiFive as they're a SiFive investor). The closest I know of (using SiFive cores) would be the Bouffalo Lab chips, but they are not just microcontrollers but WIFI/BlueTooth which of course by their very nature use considerable power. The BL602 datasheet says it uses 500 nA in "hibernate" (RTC or GPIO wakeup) and the CPU 22 mA at 192 MHz with radios off, so 115 µA/MHz. Certainly not in the league you quote. Well, actually ... I just looked and the STM32U5 datasheet says 440 nA "standby" with RTC or GPIO wakeup, so actually 440 vs 500 is not all that different, comparing apples to apples.

dragontamer 4y ago

What's the ULPMark (power benchmark) on any of those SiFive RISC-V chips? STM32U5 (one of the many ARM Cortex-M33 designs) is 464 ULPMark-CP. I don't think there's even a power benchmark on any of those chips you've mentioned. At least, I'm not seeing any online in my searches. STM32U5 (and most other Cortex-M33 based microcontrollers) are ~110nA of power consumption sleep mode, and 16uA / MHz at 160MHz, some of the most power-competitive chips I'm aware of on the market. I'm talking "run a microcontroller on a CR2032 cell for 10 years" kinda thing here. -------- EDIT: Looking at the ULPMark website (https://www.eembc.org/ulpmark/ulp-cp/scores.php https://www.eembc.org/ulpmark/ulp-cp/scores.php), it looks like the #1 chip is RSL15, which is ARM Cortex-M33, as I expected. With a ULPMark score of 1090 (higher the score, the less power the device uses). Nearly everything on the top list is Cortex-M33... Except LP5100 and XRM32UL051 are some Chinese Chips I don't understand. They got some impressive scores but I don't know what architecture they have. Nearly everything else is Cortex-M33 (except for the Ambiq, which is ARM Cortex-M4). Cortex-M0+ was the king of this a few years ago, but M33 has displaced it. We still see Cortex-M0+ chips due to much lower costs around though. So if RISC-V is around Cortex-M0+ level, I'd still be interested.

brucehoult 4y ago
"What's the ULPMark (power benchmark) on any of those SiFive RISC-V chips?" SiFive, like ARM, doesn't make chips, they license cores to people who make chips. The power characteristics are up to the chip designer -- what node they use, which "corner" of that node they use, and various other chip design features -- and not the core designer. I don't know of anyone who is putting out stand-alone microcontroller chips using SiFive cores. They've been concentrating on getting them into larger systems. For example Qualcomm said a few months ago they've shipped 650 million SoCs (i.e. Snapdragon XXX) with RISC-V cores in them (probably SiFive as they're a SiFive investor). The closest I know of (using SiFive cores) would be the Bouffalo Lab chips, but they are not just microcontrollers but WIFI/BlueTooth which of course by their very nature use considerable power. The BL602 datasheet says it uses 500 nA in "hibernate" (RTC or GPIO wakeup) and the CPU 22 mA at 192 MHz with radios off, so 115 µA/MHz. Certainly not in the league you quote. Well, actually ... I just looked and the STM32U5 datasheet says 440 nA "standby" with RTC or GPIO wakeup, so actually 440 vs 500 is not all that different, comparing apples to apples.