[Gnuk-users] Gnuk on a faster MCU

[Aurelien Jarno]

On 2017-09-12 08:06, NIIBE Yutaka wrote:
> Hello,
> 
> Another thing you can try for RSA is that...
> 
> If it has more RAM, you can change the line in
> gnuk/polarssl/library/bignum.c:
> 
>     size_t wsize = ( i > 1024 ) ? 4 : /* Because of not enough memory.  */
> 
> to, say, 6 (or 5 if not more).

Thanks for the hint. The STM32L432 has indeed more RAM, so I have been
able to use 6 as in the original code. This gives a small improvement,
I get 3.49s for RSA4096.

> If still not fast enough, I'd suggest to find a chip with 120MHz.
> 
> Yesterday, I checked NXP and found Kinetis/Freescale K22F @ 120MHz.
> 
> https://www.nxp.com/products/microcontrollers-and-processors/arm-based-processors-and-mcus/kinetis-cortex-m-mcus/k-series-performance-m4/k2x-usb/kinetis-k22-120-mhz-cost-effective-full-speed-usb-microcontrollers-mcus-based-on-arm-cortex-m4-core:K22_120
> 
> MK22FN128CAH12 - Kinetis K 32-bit MCU, ARM Cortex-M4 core, 128KB Flash, 120MHz, USB, WLCSP 64
> 
> MK22FN256CAH12 - Kinetis K 32-bit MCU, ARM Cortex-M4 core, 256KB Flash, 120MHz, USB, WLCSP 64
> MK22FN256CAP12 - Kinetis K 32-bit, ARM Cortex-M4 core, 256KB Flash, 120MHz, USB, 80WLCSP
> MK22FN256VMP12 - Kinetis K22: 120MHz Cortex-M4F Performance MCU, 256KB Flash, 48KB SRAM, Full-Speed USB, 64-MAPBGA
> 
> MK22FN512CAP12 - Kinetis K 32-bit MCU, ARM Cortex-M4 core, 512KB Flash, 120MHz, USB, WLCSP 80
> MK22FN512CBP12 - Kinetis K 32-bit, ARM Cortex-M4 core, 512KB Flash, 120MHz, USB, Thin 80WLCSP
> MK22FN512VMP12 - Kinetis K22: 120MHz Cortex-M4F Performance MCU, 512KB Flash, 128KB SRAM, Full-Speed USB, 64-MAPBGA
> 
> It includes a regulator (from USB Vbus +5V) and can run without crystal
> (no capacitive sensor, just like the chip of FS-BB48).
> 
> 0.5mm pitch MAPBGA is easier for PCB design/manufacturing/assembly.  And
> I think availability of the part is better (than other WLCSP ones).  But
> thickness is 1.23mm.  I think that it's not acceptable for the PCB
> design which uses bottom layer to put a MCU (like the design of Tomu).

I am a bit worried by the MAPBGA package, it looks very challenging to
me. Those MCU also exist in LQFP64, but in that case it's probably
better to get a faster one running at 168MHz or even 180MHz.

Also those MCU seems to only have 256B of cache, and for what I have
seen on the STM32L, the cache is important to mask the wait states and
get good performances.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien at aurel32.net                 http://www.aurel32.net