Raspberry Pi Foundation moves into microcontrollers with the $4 Pi Pico using homegrown silicon The Raspberry Pi team has announced its latest bit of hardware – the $4 microcontroller-class Raspberry Pi Pico. Some 37 million Raspberry Pi computers have been sold to date, according to head honcho Eben Upton. And although hobbyists, educators, and industry pros have seen success in connecting the diminutive computer to the …
"Help me out here, is "PIO" the same sort of thing as Arduino CCL configurable logic - a bunch of programmable logic gates that are code-configured but operate independent of the CPUs?"
I think CCL on steroids. The CCL look pretty well one shot triggers while the PIO's run some sort of microcode which makes them more flexible, while being as fast
look at this example where someone output a HDMI feed purely through software logic https://github.com/Wren6991/PicoDVI
While I truly hope RiscV continues to gain traction, when the Pi was launched the philosophy of the company made ARM the way forward. They never set out to revolutionise the IT world (so many behind this project had already done that), they wanted to reopen it to everyone.
Opening it to everyone would be the point. The original Pi didn't have to open every detail of the SOC since it was designed primarily for educating people about programming, Linux, and the like. But this is not really that kind of thing. The lack of ready-made IO makes it a different type of product than the computers they've built before, and while they undoubtedly expect some to buy them for entertainment or education, I expect they're aiming for industrial users. Especially as there are already a bunch of microcontroller-based boards which easily work for educating those new to coding; my favorites are Adafruit's more recent products.
This doesn't have to be a bad thing. The foundation may have focused on education, but they and we know a lot of Pis aren't used for that purpose. If they want to make a microcontroller that's primarily intended for industrial use, more power to them. Still, they have usually stood for openness of platform, and they designed their own chip this time. Had they used an open-source ISA, they could provide a lot of interesting details that would have served an educational purpose for those interested in the design of CPUs, while not impacting product quality. It's their decision, but I think that would have been a good one.
What precisely isn't open about the Arm-M0 ISA?
Yes, you can't make a gate-gate copy and sell it under your own name without paying ARM 0.01c, but it's not like there are secret instructions you aren't being told about which you could use on RISC-V
The original Pi wasn't totally open because they used an off-the-shelf SOC and the maker of that part doesn't let you use it how you want. This is because they sell the same part for different applications at different price points with different features enabled - that's the economics of SOCs.
It wasn't because it was only for education and kids don't need to know details of the shader architecture.
It's only a microcontroller - and the expectation is presumably that it will be used alongside a 'real' Raspberry Pi to provide those low-latency, low-power, close-to-the-metal functions that make a microcomputer running a real-time OS stumble - but I'm surprised it doesn't have WiFi (or Bluetooth, or LoRa, or at least some sort of connectivity).
I'm seeing a lot of applications for ESP8266 and ESP32 microcontrollers, rather than Atmel or Arm parts, simply because they DO provide on-package wireless connectivity. I do hope the Raspberry Pi foundation have a WiFi capable sibling of the RP2040 waiting in the wings.
Adding WiFi (in any form) would greatly increase the complexity of the chip. I guess they could license a pre-designed module for inclusion, but that's a further complexity and cost.
It would be interesting though. Using the ESP32/8266 microcontrollers is handy partly because of the wireless connectivity.
I personally use the ESP32 (WROOM32-like modules) since it has all of the above (except USB that is not needed) plus BT, more RAM, freeRTOS support, lots of GPIOs, PWMs, many ADCs, three HW serial interfaces, etc. and is not much more expensive. With very little in additional components you may also even get native (embedded PHY) Ethernet connectivity with very decent speed.
I've used the ESP8266 by the hundreds in the past, and now use the ESP32 by the thousands. My only concern is making sure I am in control of the freeRTOS and application binaries loaded, since I am not ready to rely on raw chinese OS FW.
Arduino have a R2040+WifiBlueTooth module PCB already assembled. If you check out the Raspberry Pi website you will find there are already several RP2040 based boards (somewhat?) ready.
The usual Pi process is to get something out and then build on that foundation. I am sure if these are successful we will see ones with WiFi in the future
Another point is not to assume the Pi foundation is a traditional commercial company. Its main task is education. It is not looking to compete with Arduino, ESP32 or any other product, it is creating building blocks to teach software/hardware concepts. If the product also becomes commercially successful, more the better, but it is not the primary objective
WiFi approval is long and expensive. ESP32 get round this by not bothering ...
There are official ESP32 modules from Espressif that have, e.g., FCC approval. As I understand it the approval applies to the whole module including the on-board antenna, not to the chip itself, so it applies only to those modules and not to all/any ESP32 boards ... but by the same token the approval will apply to any product that uses such a board (without altering its radio characteristics).
That's why you get products like the Adafruit Huzzah ESP32 breakout that has a certified WROOM32 module (complete with antenna) mounted on it, and so benefits from the WROOM32's certification.
I would have liked to see more analog ports. That would have been my main interest in this. At 12bit resolution, that's almost good enough for generating MIDI NRPNs - mind you, a lot of systems only actually resolve 10bit. Three ports gets you one joystick and one foot pedal.
I would also have liked USB2, although I suppose 1.1 would again be good enough for standard MIDI.
Analog multiplexing is the usual approach, as you rarely need to sample anywhere near the maximum sample rate. If you're reading human input then you're very unlikely to require anything more than 40-50Hz - for a lot of purposes, even a 20Hz sample rate may be acceptable as people don't move that fast.
The RP2040 appears to be rated at 500,000 samples/sec. If you average 100 raw samples for noise immunity, that's 5000 samples/sec. Ignoring switching and settling time you could multiplex 1000 50Hz human inputs per ADC channel. Even if you lose 90% for switching and settling time, that's still 100 joystick axes...
A 4051 analog mux chip (40p) has two channels of 4:1, bigger ones are a tad more expensive. Three of those gives your joystick up to 15 axes - probably enough for most purposes.
The analogue is poor on this. Despite it's apparent 12 bit range, the datasheet says it's only 9.5 effective number of bits. It needs a 48MHz clock and then takes 96 cycles to actually do the conversion. Typically you can get a 12bit SAR conversion to run in 14 cycles. One for the the sample and hold, twelve for the conversion and the final one to move the result to the output register.
The temperature range is only -20 to +85 rather than the more standard -40 to +85 of industrial products.
Finally, why on earth have they used a Cortex M0+ and then added an eight cycle divider as a peripheral? I see they selected the single cycle multiplier for the core itself.
Why not just use a single Cortex-M3 with its single cycle multiply and divide, then you get all the MACC instructions too.
I suspect that this would have made it very like all the other Cortex-M3 devices out there and the dual M0+ does at least give it a differentiator.
An "ideal" ADC’s SNR: SNR = 6.02 × N + 1.76 dB, where N is the ADC’s resolution in bits.
If you look up a a low noise, low power, high performance 32-bit ADC e.g. LTC2500-32 (€53.50) and calculate the ENOB from the stated SNR given in the datasheet you may be shocked:
i.e.
From the datasheet for the LTC2500-32: 104dB SNR (Typ) at 1Msps
ENOB = (SNR - 1.76)/6.02 dB
ENOB = (104 - 1.76)/6.02 dB ~= 16.98-bits ~=17-bit
Lets pick another one say a 16-bit ADC e.g. AD9656BCPZ-125 (€300.77)
From the datasheet:
SNR = 79.9 dBFS at 16 MHz (VREF = 1.4 V)
SNR = 78.1 dBFS at 64 MHz (VREF = 1.4 V)
ENOB = (79.9 - 1.76)/6.02 dB ~= 12.27 bits ~=13-bit
I do not see anything wrong with with an ENOB 9.5 for a 12-bit ADC.
This post has been deleted by its author
So this is not a little Linux board. It isn't an Arduino or ESP32 either. Neither is it compatible with Arduino libraries. It needs a brand new IDE by all accounts. The world is already running on any of the thousands of STMicro M0 boards out there.
So it is difficult to see what this addition to a crowded market has to offer - apart from the "Pi" branding.
Hobbyists are already well catered for and industrial users buy the chips by the million to integrate into their own devices (with the help of STM development tools).
Uses VSCode for C, although you don't need to use an IDE if you don't want to. Command line compile and drag and drop works fine. Eclipse also works.
There are a number of (mostly) unique features. Twin cores, PIO, HW PWM, HW interpolators, sophisticated DMA, HW Dividers. All combines to give the ability, for example to drive two DVI displays...
Great, buy one of them instead.
Pi is not competing against other boards, but expanding its range of tools for education. I am pretty sure no one in the Pi foundation said lets destroy the opposition with this baby..
If you look at the regular Pi boards, every month a new "pi" killer comes out, faster, better specced etc. However what they don't have is the foundation support, tools and education services that allow people programming for the 1st time to quickly get up and running. That is the pi foundation core market. People, especially young adults, with little programming experience who want to connect computers to the real world or want to learn to program in a cheap environment
"Pi is not competing against other boards, but expanding its range of tools for education."
That is a genuinely noble goal, and obviously they do sell like fairly hot pies, but it's so very much the "British" way. If it sells quite well, we're happy, if it doesn't, we grumble a little, and do something else.
Part of me does think: wouldn't it be nice if they really did try and launch some really slick devices as well that could really sell well, so that they could make quite a tidy bit of extra cash (and use more of it to drive development and the core mission forward too).
Pi's are good at what they do, but to think of just one possible example: I'm currently using a Pi as a home fileserver (with an external USB drive attached). It works not badly, and it was quite fun finding out how to make it work (part of the attraction, of course). But it's running off an SD card, in a slightly awkward case which has ports and cables coming out two different sides of it, and with a rather awkward Micro-HDMI cable/connector, too.
The fact that it is all pretty much as small and minimal as you can get is obviously good for a lot of "hacker" type experimenting, but something with a little more polish would be nice too. What I'd really like (for my use case, and I'm sure it's not an entirely uncomon one, not everyone is wiring things up to the GPIO header, neat as that is) would be some sort of Pi-Plus: all the connectors coming tidily out just one side, a hardware clock on board, and one, if not two, SSD slots, so that it's running off something a bit more reliable than an SD card. That would be a nice mini home server, and while, yes, it would cost a bit more than a typical RPi kit, it would still be good value for money, but could bring in quite a bit of extra income, which would help in the long run.
"[...] with a rather awkward Micro-HDMI cable/connector, too."
I bought a Pi Zero and its official case and cables. It was disappointing that the HDMI connector wouldn't lock into the board due to the case thickness. In fact it didn't even connect most of the time. Another supplier's HDMI cable would stay in - most of the time - but also wouldn't lock. Without the case the connectors locked ok.
I complained and was told to return it. However Google showed that much earlier reviews had also highlighted this incompatibility on what was supposed to be a Pi Zero custom case. The connector is now tightly bound to the case with sticky tape.
I have to question this, since there are already several microcontroller-based education projects. I wonder if education is the foundation's primary goal with this board. When the original Pi came out, there was little competition in SBCs running Linux on low-cost components. Today, there are a lot of boards like it but as you said, the Raspberry Pi is still the one with all the tutorials and support. It therefore makes a lot of sense for the foundation to continue to create boards like that for the educational benefits.
In microcontrollers, however, there are a bunch of existing educational boards. From the early Arduinos to the MicroPython-focused ones to that weird thing the BBC made. I haven't yet seen what this does for education that those boards don't do. For those new to programming, most of those boards will be better since they include peripherals. An early programmer can blink an LED that's already on the board, then move on to breadboarding on some more. Or they can make a board that uses hardware which either comes on the PCB or can be attached conveniently so they get used to the peculiarities of driving hardware manually. The alternatives are usually much harder for someone new to programming, as they also have to learn to assemble the hardware and determine whether something is not working because the connection isn't stable or the code's wrong.
For this reason, I doubt this is for the same educational goals that the larger Pis are for. That doesn't have to be the foundation's only goal though, so this isn't necessarily a problem. Still, arguments that cite it as an educational board don't make much sense to me until I hear why it does the job better than the many tested alternatives in that space.
VS Code is already used in embedding, e.g. the PlatformIO extension has support for a bunch of embedding backends including Arduino. There is also a dedicated Arduino extension if you prefer.
VS Code is becoming popular because the Arduino IDE is a bit cack. It does the job if you have a sketch you need to compile+install but if you're writing code, then editor doesn't have the niceties you'd see in another IDE, e.g. ctrl+click to see where a value is a defined, or function auto complete.
I think the Pi Pico should have an Arduino compatible or look-alike API at least as an option on the side. Simply for code portability if for no other reason.
That's true too. I have a box of bits & pieces for my Arduino - temperature sensors, controllers, LEDs etc. and they all have libs to make it easy to make them do what they're supposed to.
Even the likes of Teensy which is an ARM based embedded board has a compatibility API. It seems odd that the Pico doesn't because they seem VERY price competitive but if it means rewriting stuff then that advantage could be gone.
It's the API rather than the pins. Most sketches / libs will put define the pins as constants or params to make them easy to change. The actual APIs are different which would necessitate completely rewriting any existing code or libs rather than utilising the existing ecosystem.
I'm sure someone will produce a compatibility layer for Pico, but it's surprising it wasn't one in the first place.
Excellent example to prove the point you argue against. The Model T is well-known because it was affordable due to efficient manufacturing. Not unlike the original Pi itself. It didn't necessarily do more things, but it had a serious point of competition where it provided a benefit to the consumer over previous cars. Other cars came out with different features, increased reliability, or even more affordable, and they succeeded. Usually, they needed some point of competitiveness in order to be successful cars.
The original question is just asking what this board's competitive edge is. It could be anything: "it does things the others don't do", "it does things the others do at less power", "it does the same things but it costs less", "it uses the same power but is faster", anything like that. And some of the advantages sound a little interesting, though perhaps not as differentiating as I'd hope. If it didn't have any advantages, then it'd be like many other cars you've never heard of, because everyone looked at them and decided to buy the Model T instead. It's not irrational to ask what the differentiating factors are if they're not evident.
So it is difficult to see what this addition to a crowded market has to offer - apart from the "Pi" branding.
The sole extent of my knowledge on the Pico is this article, so far. But it appears to be a mid-point between rPi and NodeMCU type boards like ESP32. I still have some questions about code deploy, supporting libraries, and communication options. But I can see it falling in an area where ESP32 doesn't have enough programming/processing capability but running a full blown OS (Linux) is too fat.
"I can see it falling in an area where ESP32 doesn't have enough programming/processing capability but running a full blown OS (Linux) is too fat."
That's not where it is. This has two M0 cores at 133 MHz. The ESP32 has one or two cores running at 160 or 240 MHz with extra acceleration for some tasks, especially cryptographic ones. It also has double the memory. This has to fall into a place where an ESP32 is too much or doesn't have the needed features.
>So it is difficult to see what this addition to a crowded market has to offer - apart from the "Pi" branding.
So it's an Arduino Due with RPi built-in "educational approval" and hopefully a well supported infrastructure for the price of a Chinse Arduino-Nano knock off ?
Bubblegum micro that's not terrible in technical terms (nice speed, lots of RAM but external flash?) and a dev environment that looks to be targeted at Pi only (so far), no need for J-TAG adapters etc
Hmm, curate's egg....
Documentation looks great and I don't think it'll be long before the IDE goes cross platform, the price is good for what you get, yeah, I've ordered a couple and will tinker while watching with interest.
264kB of RAM? Makes my life harder, lots of juggling needed to marry decent internet cryptography/security with the needs of my application code. 1MB would feel very spacious in comparison. Is it asking for 1MB RAM too much? ESP32 is better (520kB, but significant fraction used by the freeRTOS) but still limited. PSRAM does not count.
I started on systems that had 256 BYTES of RAM and managed to get them to do a lot...
The problem with struggling with small amounts of RAM is that your project ends up being about squeezing your memory usage, not about what you really want your project to do in the first place.
In the past devices were so constrained by memory that it often took up a vast amounts of programmer time and needed specialist tools like overlaying linkers, but now hardware is cheap.
I recently struggled myself to get sql INSERT to a remote server running on an Arduino Nano + Ethernet board alongside other code because I hit the ATmega328p 2 KB RAM limit. After a day or so of trying to fix it, I decided that it was like learning to fart the "Yellow Rose of Texas" - possible, but really not worth the trouble.
"but now hardware is cheap"
Cheapness is relative, based on what your project actually is.
A one-off that you just need to get up and running ASAP - grab a high-end off the shelf devboard, make use of whatever pre-written code libraries are available, bish bash bosh, get it working and don't worry about how woefully inefficient it all is in terms of materials costs, memory utilisation etc.
A design intended to go into long term/high volume production - now you need to be a bit more cautious about balancing development costs against materials costs, because a) you could end up with a design that costs so much to manufacture that it wouldn't be profitable to sell at the required price point, and b) even if it is profitable, how much *more* profitable could it have been if you'd spent an extra few months optimising the design to reduce materials costs?
"[...] how much *more* profitable could it have been if you'd spent an extra few months optimising the design to reduce materials costs?
There was an obvious way to upgrade one of our network products that meant buying Intel's new expensive 386 cpu. A problem was that it would require creating a slightly larger box than our standard one. The new certifications would take time and be expensive.
I had been dabbling with the new Xilinx FPGA's in a way that could be used to achieve the existing unit's required feature boost. It was a risk - the hand-crafted routing used nearly every gate on the chip - but it worked. The competition had all chosen the expensive 386 path - thus setting a market selling price which gave us an enviable margin.
Agree with the scentiment (similar experiance with 8031/8051 devices in the past) - but if every chip came with 8GB (RPi4 max) then the cost of most embedded systems would prohibit actually implementing them.
From a software view that sounds like normal embedded development (or any to be honest), hitting a limits of hardware resources, trying to workaround the limitation in someway and then at some point (after an hour or a year of effort) deciding that the problem is better solved using different hardware - may only be the next chip up in the line.
Perhaps hardware is spec'd correctly for the original requirements - and allowed the cost of the device to cost only £1, allowing millions of the product to be sold. Adding another feature might only need an extra bit of RAM or an extra byte of instruction ROM, but if hardware doesn't have it then it won't be possible without changing the hardware platform. Upgrading the hardware platform might cause the platform cost to increase to £2, reducing the product viability. This is not an uncommon problem.
If millions more of the product are sold then it might be worth the developer time to maintain RAM/ROM optimised code. If not, then revise hardware and sell version 2 of the product with the extra feature + cost.
RAM is "cheap" as a stick for adding to a PC, in the embedded world it is not cheap compared to the overall manufactured part.
In your example either the platform is under spec'd for the intended use or a resource limit has been reached as extra features have been added.
This post has been deleted by its author
256 BYTES! We used to program with 8 BITS of RAM! Every time we rewrote the byte we had to scrawl a letter on the cardboard box we lived in, in the middle of the road...
(8 BITS and a cardboard box! We just had ONE BIT! It was a hole in the road and when it filled with water it was 1 until we bailed it out to make it 0 ...)
[Sorry, couldn't resist. Programming in 256 bytes is quite a skill.]
Indeed. The most resource constrained micro I've worked on so far (and hopefully ever) in my career had 32 bytes of "RAM" provided by the 32x8-bit GP registers in the core itself.
Mind you, even the relative behemoths I'm working on right now still have less memory than my first 16-bit home computer, though they make up for it by delivering a crapton more processing power - 128KB SRAM attached to a 170MHz Cortex-M4 core, vs 512KB DRAM attached to a 7.14MHz 68000 - so all of those memory saving tricks I learned back in those heady days of 80's home computing when programming was just something I did for fun of an evening still come in handy all these years later...
It also helps to remember that, whilst some embedded developers do need microcontrollers with specifications that wouldn't disgrace a half-decent desktop PC, there are many more projects out there (both hobbyist and commercial) where such performance would be overkill, in some cases massively so - so whilst this module might not tick all the boxes for some, it will surely tick the boxes for more than enough others to justify its existence in an already reasonably crowded market.
This post has been deleted by its author
That's very common for embedded devices and it can still do a lot. e.g. until recently most 3D printers were using 8-bit boards with 128Kb of RAM. By embedding standards the Pico is quite good especially at that price.
I'm sure crypto could be done in the space too although it would be better done by hardware in the chip. There are probably embedded boards that do it too if the Pico is no good.
Dual Cortex-M0+, with a gobload of SRAM? That is rather curious juxtaposition of both low-end (M0+) and high-end. That much SRAM would also make the low-power nature of the M0+ less relevant. And two M0+ cores, when something like an M4 + M0 would have been much more interesting from a performance & low-power perspective.
Most of my Cortex-M work revolves around the STM32, specifically the F0 (often <16 kB SRAM), as well as the F4 (M4-based) and F7 (M7-based), with most recently the H7 (M7 + M4 core).
An annoying quibble with this breakout board they have is that the SWD header only has three pins (SWCLK, SWIO & GND), when 3.3V is also commonly added to the SWD header. Means no easy use of standard STLink and similar plugs.
If RPi Foundation wants to take on Microchip, ST and other MCU manufacturers with this chip, I wish them luck, but I'm honestly not sure what market they are targeting with it.
> I'm honestly not sure what market they are targeting with it.
Quite.
ISTM to release a small device like this that does not have WiFi is just plain dumb. They have ruled themselves out of the IoT field and as such this is just another dev board
As for the $4 price tag: it's too low for retailers to make any worthwhile profit from. Although I suppose it will go the way of the $5 Pi-Zero which is either unobtainable as the margin makes it not worth stocking, limited to ordering 1 otherwise it would cannibalise the more expensive products' market or is overpriced so (including P&P) as to be uncompetitive.
Whilst IoT is probably of the biggest markets when it comes to what is "interesting", there are a lot of others where WiFi (and other radio interfaces) are of no interest at all - such as industrial automation (sensors and actuators) and automotive, where things like CAN, Flexray and Ethernet are more useful (and which are also not included).
However, I do think this device will be useful for kids and anyone else wanting to learn the principles of embedded systems without the need to trawl through a very large (1500 page+) manual that attempts to explain how everything fits together.
Quite. Just look at the number of different Nucleo devboards that ST have released - most of those have next to nothing in the way of onboard connectivity bar the USB socket for programming/debug, plus a USART or two exposed via the header pins, and yet they're phenomenally popular both with hobbyists wanting to learn a bit of ARM-based development, and with professionals wanting an off the shelf "compute module" that can easily be integrated into another bit of hardware.
If nothing else, I think releasing modules like this helps to show up how much of what goes on in the embedded systems world is so well hidden that it doesn't register with the general public at all - everyone and their dog has IOT on the brain thanks to all the media coverage it gets, but for every IOT-enabled device out there, there's countless more doing less glamorous bits of work. Even just within the PCs that some of us are using to read this site there'll likely be a good handful of embedded systems doing their own thing - display driver, SSD controller, touchpad controller etc - in the background without being given a second thought.
They're not knockoffs because the design is open source. I have an Elegoo clone which came with a big kit & no complaints so far - it is well made and works. Some of the clones are stupidly cheap on Amazon which is actually a selling point for the Arduino platform - tinker with the cheap stuff but you have the peace of mind of the branded stuff if you need it.
Wasn't sure about this at first But the Pi foundation does a few things very, very well - primarily documenting stuff and removing the initial barriers to adoption. I've got a bit bogged down migrating from Atmel to Arm because it looked way more complicated, but I suspect this will change that. I am quite partial to Wifi access though. Maybe for v2 - the pi foundation has already paid for that fancy triangle-shaped piece of nothing that serves as an antenna, may as well get some use out of it ;-)
(edit - I see the Nano RP2040 Compact at https://www.raspberrypi.org/blog/raspberry-pi-silicon-pico-now-on-sale/ already does - no doubt there will be more to come). (edit 2 - actually the PIO looks pretty cool too. People are going to do some fun things with that)
To be fair, the guy who built the dual DVI system was also the guy who defined the PIO's instruction set.
But yes, I've seen a lot of interest in the PIO. A few other devices have something like it - the beaglebone's PRUs, Motorola's TPUs, the Cypress USB engine, the parallax propellor .. but they're largely seen as rather specialised, for the expert. These cheap and accessible PIOs that can be used to generate video could get a bit more interest going in assembler programming and FPGA state machines.
There is always a trade off there, A/D is expensive. You'd be looking at some combination of piss poor sampling rates, "I don't think I'll bother" pricing and low order bits than are more random number generator a than actual resolution. Every few years I find myself shopping for ADCs, the prices of even 12bit @ 1MHz are enough to make you wince when the magic smoke gets loose, they're not general toys you have knocking about for random experiments. If you want a better ADC, spend the £20, £30, £100 on a chip with the right capabilities and wire it up.
True 16bit are still expensive, because you need reference voltages that are good to much better than 1/64K. The ones you can afford are successive approximation (ie don't believe the last 4 digits).
I remember when Burr-Brown 16Bit at 100Khz were a thousand quid and came in their own metal box.
This post has been deleted by its author
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
