Teardown of the Pi Pico 2 to the silicon level. A look at the RP2350 die, the Richtek RT6150 die, and the QSPI die. Link to the photos and blog: electronupdate... silicon die, analysis
The criticism of microusb is that for one the cost difference is shrinking to almost nothing; it's more convenient due to being reversible, and eliminating microusb is more convenient that people don't need to use different cables again. I think they're only using microusb for full footprint compatibility and because it's in their parts bin - switching to USB-C wasn't yet feasible 3 years ago. ARM microcontrollers often include internal linear regulators, to reduce external part count, so they only need to be supplied with IO voltage (3v3) which is shared with the bulk of the system which integrates the microcontroller, while a lower voltage rail is internal to the core. If Pi Pico a board for hobbyists was the only or primary product, they wouldn't have done that, but the microcontroller is supposed to be sold to electronics manufacturers to integrate into real consumer-market devices, which can be sensitive to part count, layout requirements and overall footprint. I think ARM is potentially going away in the future of this microcontroller lineage to be replaced by RISC-V due to licensing costs. The new cores are already included, just waiting for ecosystem maturity.
Awesome! We just got some STM32Us here which are on ST's new Samsung 18nm. Wish I could see the die differences. Pretty nice leap in power consumption.
The voltage regulator is to provide the 1.3v core voltage. I understand it's mostly there to simplify designs using the RP2350, removing the need for an additional voltage regulator. The minimal viable RP2350 design is little more than a 3.3v regulator, the RP2350, the qspi flash chip, an inductor, some caps and some resistors.
Excellent look into the silicon as always! The reason why they have a buck-boost PMIC is because they intend for you to be able to connect VSYS directly to a battery which may be under 3.3 V. Really wish they used USB C, the cost differences would probably 1 or 2 cents for the non-USB 3 USB C connector and the two CC resistors. Even when they were designing the original Pico it was feasible. I can only assumed they were trying to compete with the Arduino Micro so just copied the micro USB. As a result I see lots of applications that just ignore the micro USB connector and have a separate USB C connector for power on their own PCB or don't use the Pico module at all and just the DFN IC.
The voltage regulator is integrated with the power saving circuitry - you can configure it to automatically change from switching to linear mode when entering sleep. This way you get both low quiescent current and good efficiency. That said, I doubt it will be used on many products. A discrete LDO will take up less board space, have lower BOM cost and be easier to certify for EMI (the reference hardware design is very particular about both layout and inductor part number, similar to the crystal oscillator). Also, footprint compatibility be damned, lack of USB C on the pico 2 is inexcusable.
the metal fill is not really for etching purposes. It is to help CMP. Without metal fill, there would be large within-die variation. Metal-fill mitigates this.
The voltage reg on there is a buck converter down to 1.1V which is the core voltage. Apparently the components for that circuit on the PCB are fairly precisely specified, seems the on-die regulator is very very picky for stable operation. (It apparently cares which direction the inductor is wound, etc.)
Ppl are grateful for die images, but i wanted to thank you for running a blog. Not so common nowadays to have something valuable to read and hosted elsewhere besided big tech
Great video! The RP1 is the support chip on the Pi 5 (and it started being designed before the original RP2040 in the Pico) - so that with the Pico and Pico 2 accounts for three RP variants - wonder what the other one is/was?
Nice work on this! How do you remove that top metal lower, i.e. that power distribution network? I assume chemicals so you don't ruin anything underneath it?
Seems to me they did microUSB simply to be completely 1-1 compatible with the previous Pico. Otherwise surely the cost differences have to be getting into the low cents range, which we're already paying for the new v2 chip.
Hi Electronupdate, we were discussing the RPi on the SiloconPr0n discord managed my John McMaster. While there isn't much you've missed in your analysis, Lennert posted a partial annotated IR backside image which might interest you. You've pointed to the UL marking on the board. These markings are for the PCB itself, not the whole product. E.g. the boards are from FR4 and complies with standard UL94V-0. The CE and FCC logo do say something about the whole board, including radiative emissions testing and ROHS. While these aren't a big deal for a board without radio, I always found it odd these markings are applied very early in the design and manufacturing stage, usually long before you can say it passes the testing. (Some fields are very picky about this, and even a simple silkscreen change is not allowed.) As for the current bug bounty, the magic value is programmed into OTP and isn't in ROM. But ROM could still be interesting to read out to find a bug in the secure boot.
there is always going to be the next raspberry pi , but i think its way far ahead , we currently have RPI with 8 gigs of ram and pretty powerful chip , and RPI is known for having massive increments of performance from the previous gen , i dont think they have that yet ,the pi 5 is still in its beginning stages ,to get a massive uplift i think we have to wait for 2026 (maybe) to get a compleatly new chip ,in between they will revise the design and we would probably have PI 5 with better connectivity or better ram speeds
My question is that would this Raspberry Pi chip that costs basically pennies to make, is more powerful than a Pentium 200 MMX from a white ago ? I mean raw performance because if it is, it is incredible how far we got.
A lot of it comes down to die size and process node. The Pentium was using the smallest node available at the time, and still it ended up being a big and costly die. The pi chip is using a 10 year old process node so it’s cheep. Plus the die is super small so you can fit tons of them on a single wafer.
Thank you for the excellent images!
Would love to see higher resolution of the sea of gates to try to sus out how much is the ARM cores and how much is the RISC-V.
Yes, but that's not possible with a optical microscope.
You always disappear but come back when something shiny comes around!
The criticism of microusb is that for one the cost difference is shrinking to almost nothing; it's more convenient due to being reversible, and eliminating microusb is more convenient that people don't need to use different cables again. I think they're only using microusb for full footprint compatibility and because it's in their parts bin - switching to USB-C wasn't yet feasible 3 years ago.
ARM microcontrollers often include internal linear regulators, to reduce external part count, so they only need to be supplied with IO voltage (3v3) which is shared with the bulk of the system which integrates the microcontroller, while a lower voltage rail is internal to the core. If Pi Pico a board for hobbyists was the only or primary product, they wouldn't have done that, but the microcontroller is supposed to be sold to electronics manufacturers to integrate into real consumer-market devices, which can be sensitive to part count, layout requirements and overall footprint.
I think ARM is potentially going away in the future of this microcontroller lineage to be replaced by RISC-V due to licensing costs. The new cores are already included, just waiting for ecosystem maturity.
Personally I have switched to only USB-C cables, I now use USB-C to Micro-B adapters where needed - that has worked well for me.
The PCIe IO controller chip on the raspberry pi 5 is the RP1, so if they are at RP4 with the RP235*, only one is missing in the numbering.
The RP3 is the RP3A0 on the Pi Zero 2W
Awesome! We just got some STM32Us here which are on ST's new Samsung 18nm. Wish I could see the die differences. Pretty nice leap in power consumption.
Absolutely love your videos, thanks so much for making them!
The voltage regulator is to provide the 1.3v core voltage. I understand it's mostly there to simplify designs using the RP2350, removing the need for an additional voltage regulator. The minimal viable RP2350 design is little more than a 3.3v regulator, the RP2350, the qspi flash chip, an inductor, some caps and some resistors.
Many modern MCUs come with built-in voltage regulators, like the STM32H7 series.
Excellent look into the silicon as always!
The reason why they have a buck-boost PMIC is because they intend for you to be able to connect VSYS directly to a battery which may be under 3.3 V.
Really wish they used USB C, the cost differences would probably 1 or 2 cents for the non-USB 3 USB C connector and the two CC resistors. Even when they were designing the original Pico it was feasible. I can only assumed they were trying to compete with the Arduino Micro so just copied the micro USB. As a result I see lots of applications that just ignore the micro USB connector and have a separate USB C connector for power on their own PCB or don't use the Pico module at all and just the DFN IC.
Always appreciate your videos, thank you
The voltage regulator is integrated with the power saving circuitry - you can configure it to automatically change from switching to linear mode when entering sleep. This way you get both low quiescent current and good efficiency.
That said, I doubt it will be used on many products. A discrete LDO will take up less board space, have lower BOM cost and be easier to certify for EMI (the reference hardware design is very particular about both layout and inductor part number, similar to the crystal oscillator).
Also, footprint compatibility be damned, lack of USB C on the pico 2 is inexcusable.
the metal fill is not really for etching purposes. It is to help CMP. Without metal fill, there would be large within-die variation. Metal-fill mitigates this.
What is CMP?
@@JohnLauerGplus Chemical Mechanical Polishing. See wikipedia for details.
Thank you very much very interesting!
you're an amazing lad, great youtuber, seemling quite knowledgeable enginerd; thats a new mainstay
The voltage reg on there is a buck converter down to 1.1V which is the core voltage. Apparently the components for that circuit on the PCB are fairly precisely specified, seems the on-die regulator is very very picky for stable operation. (It apparently cares which direction the inductor is wound, etc.)
Ppl are grateful for die images, but i wanted to thank you for running a blog. Not so common nowadays to have something valuable to read and hosted elsewhere besided big tech
Great video! The RP1 is the support chip on the Pi 5 (and it started being designed before the original RP2040 in the Pico) - so that with the Pico and Pico 2 accounts for three RP variants - wonder what the other one is/was?
The RP3A0 is the SoC for the Zero 2W
Thanks very much for the upload! 👍
The challenge secret isn't in mask ROM, it's in OTP, specifically to prevent reading it from the mask.
Can't remember the exact order, but this one is "RP4" because the other ones are RP2040, Pi Zero 2W chip (SoC+RAM) and IO chip on Pi 5
What a treat on a Sunday evening
Nice work on this! How do you remove that top metal lower, i.e. that power distribution network? I assume chemicals so you don't ruin anything underneath it?
great to see you again!
Seems to me they did microUSB simply to be completely 1-1 compatible with the previous Pico. Otherwise surely the cost differences have to be getting into the low cents range, which we're already paying for the new v2 chip.
Hi Electronupdate, we were discussing the RPi on the SiloconPr0n discord managed my John McMaster. While there isn't much you've missed in your analysis, Lennert posted a partial annotated IR backside image which might interest you.
You've pointed to the UL marking on the board. These markings are for the PCB itself, not the whole product. E.g. the boards are from FR4 and complies with standard UL94V-0. The CE and FCC logo do say something about the whole board, including radiative emissions testing and ROHS. While these aren't a big deal for a board without radio, I always found it odd these markings are applied very early in the design and manufacturing stage, usually long before you can say it passes the testing. (Some fields are very picky about this, and even a simple silkscreen change is not allowed.)
As for the current bug bounty, the magic value is programmed into OTP and isn't in ROM. But ROM could still be interesting to read out to find a bug in the secure boot.
Actually the boot rom is open source and you can take a look at the code and find those bugs :) So cool
Great video, love it! Just please fix your sound please. 20$ mic sounds better then this
Hey, would you be willing to take some pictures of the xbox cpu for the community? I know you did the gpu already, but cpu is needed.
2:18 if The Raspberry Pi 5 features a RP1 microcontroller, this can only mean a Raspberry Pi 6 is confirmed and launching soon :D
there is always going to be the next raspberry pi , but i think its way far ahead , we currently have RPI with 8 gigs of ram and pretty powerful chip , and RPI is known for having massive increments of performance from the previous gen , i dont think they have that yet ,the pi 5 is still in its beginning stages ,to get a massive uplift i think we have to wait for 2026 (maybe) to get a compleatly new chip ,in between they will revise the design and we would probably have PI 5 with better connectivity or better ram speeds
@@fouzaialaa7962 Raspberry Pi 5 is less than a year old so we won't see 6 till 2026 or 2027 maybe. We're waiting on CM 5 to be released.
@electronupdate do you have any projects in mind for these new Pico 2s?
It's "Made by Raspberry" or "Designed by Raspberry"?
They don't have a foundry, right?
My question is that would this Raspberry Pi chip that costs basically pennies to make, is more powerful than a Pentium 200 MMX from a white ago ? I mean raw performance because if it is, it is incredible how far we got.
A lot of it comes down to die size and process node. The Pentium was using the smallest node available at the time, and still it ended up being a big and costly die. The pi chip is using a 10 year old process node so it’s cheep. Plus the die is super small so you can fit tons of them on a single wafer.
@@BGTech1 having so much memory on board help a whole lot too.
@@cj09beira Yup! At small process nodes it can fit on the die. No need for expensive DRAM chips and circuit board space.