#687 – The RP2350 with the Raspberry Pi Team


Episode Artwork
1.0x
0% played 00:00 00:00
Jan 28 2025 95 mins   432 2 0


Welcome James Adams, Chris Boross, Liam Fraser, and Luke Wren!



  • The last time the RPi team was on the show was about the RP1 (#648)

  • The order of parts being released was RP2040->RP1->RP2350

  • Check out the datasheet for the RP2350

  • Learning from silicon

  • Security and power states

  • The part is a “Dual dual core”

  • The Arm side is a Dual M33

  • The RISC V side is a Hazard 3 processor, designed by Luke based on a previous processor called the Hazard 5

  • HB5

  • There is a mux on the core and you select which side you’re going to use at boot

  • There are 48 GPIO (but users always want more)

  • Chris Boross (first time on the show) is on the commercial team. He’s seing interesting applications for the RP2350 including devices that are using it for motor control.

  • They also have seen the part used in satellites because mRAM or masked ROM is less susceptible to radiation errors

  • The PIOs have changed, but are more evolutionary from the RP2040

  • The PIO allows you to create state machines that process inputs without processor interventions, basically like tiny cores

  • 2 cores – 8 total

  • Interesting PIO applications


  • The core frequency only increased 133 MHz -> 150 MHz. There is tougher timing with the M33

  • LVT – lower voltage threshhold

  • 30 -> 40 pins

  • There are now variants listed on the RP2350 product page (but not in mass production) that include flash in the SOM package

  • RP2040 was one power domain

  • “Powerman” (and of course AVR Man)

  • Switched core

  • AON – always on

  • 32 kHz

  • There is a C/C++ SDK that is the basis for other ports

  • Security is a focus for the RP2350

  • Bootrom in every chip

  • Secure boot

  • M33 features – secure / non-secure

  • RISC V PMP

  • RCP – Redundancy Coprocessor

  • Raspberry Pi had a challenge / bounty for getting the secret out of the RP2350 OTP with secure boot

  • One of the few silicon companies doing this sort of thing in public

  • Past guest Aedan Cullen was one of the hacks called “Hazardous threes”. He gave a talk about it at 38C3

  • Past guest Colin O’Flynn was also mentioned because collaboration around side channel attacks with the Chip Whisperer

  • IOActive used a FIB – Fine Ion Beam – and passive voltage contrast to capture an impressive image of a decapped chip (see the RPi post)

  • “Never want to see ‘novel technique’ in an email”

  • Improving the RP2350 silicon

  • How do you decide what to fix/leave?

  • Can it be changed in metal/vias?

  • SIO spinlock not being fixed

  • Chicken Bit

  • Filler cells are reprogrammable and help with fixes

  • It costs approximately $50K per layer to change (ostensibly because of the high costs of masks)

  • ULA – uncommitted logic array

  • Die shrink doesn’t seem to make sense

  • Will keep making each chip as long as 40 nm fabs are around

  • Thinking about the RP2040

  • The easiest way to get started is to use a Pico (RP2040) or a Pico 2 (RP2350). Both have connectivity options as well.

  • Raspberry Pi is now a public company! Doesn’t change much other than the business scrutiny.