Raspberry Pi has just announced that they’ll be selling their RP2040 microcontroller chips by the reel, directly to you, at a decent discount.
About a year ago, Raspberry Pi released its first piece of custom silicon, the RP2040 microcontroller. They’ve have been selling these chips in bulk to selected customers directly, but have decided to open up the same deals to the general public. If you’re looking for 500 chips or more, you can cut out the middleman and save some serious dough.
Because the RP2040 was a clean-slate design, it uses a relatively modern production process that yields many more processors per silicon wafer, and it has been essentially spared from the chip crisis of 2020-2021. According to CEO Eben Upton, they’ve sold 1.5 million in a year, and have wafers in stock for 20 million more. You do the math, but unless you’re predicting the chip shortage to last in excess of 12 years, they’re looking good.
https://direct.raspberrypi.com/
Reels of 500 RP2040 chips, with a unit price of $0.80
Reels of 3,400 RP2040 chips, with a unit price of $0.70
They are $1.00 each through distribution.
But the Pico at $4.00 each is probably more useful to the typical hacker.
Adafruit will buy the reels for their boards.
Cool. At 1.7 mm square, that’s about 10k chips on a 200mm wafer, so they got a run of around 2000 wafers. A stack about 1.5m high, 100 kg. And worth, literally, its weight in gold.
TSMC 40nm, so 300mm wafer.
100kg is about $5.8M at today’s gold prices. 20M chips at $0.5 a pop (I’m guessing here as to die pricing) is just short of double that. I.e. silicon with sh*t on it is yet again more valuable than gold.
Value of the end product isn’t all that counts, don’t ignore the fact that you need very specialized equipment to “make” this gold, use it, maintain it. Where as in ordinary gold you’ll only need and old man, a donky a shuffle and a pan. In other words, the investments and overhead greatly reduces the clean profits.
Turning sand into gold is modern alchemy.
But in 20 years time a bucket full of the yellow stuff is probably still worth something while a bucket full of those dies have returned to the cost of sand.
I was thinking of using RP2040 for a new design, but discovered they have bug in the ADC, which makes it much less accurate.
Also need to produce ships with build-in flash. External flash for MCU is PITA.
Hmm, yes – ships with flash. Damn auto correct.
External flash is a PITA, but it’s how they were able to get such a small die size, which is what allows them to both be in stock and be so cheap.
Could package the flash dies in the same package, like the GD32 clones.
And with external flash, one can size it to the end application far more easily.
And for Raspberry PI’s case, they can make more models far more easily.
Internal flash at 40nm is… problematic, not that it doesn’t work, but it’s completely different from 90nm one that you see on most old generation microcontrollers (like STM32F4 old).
For example first 40nm flash was from renesas in 2011, TSCM 40nm flash only debut in 2018.
IP for 40nm flash controller are not cheap, and know-how of this technology neither. It’s perferctly understandable that they do not want that for a first design, or a second one.
What’s the problem of using an external flash?
I hadn’t heard about this ADC bug… do you have any more information?
How are they cutting out the middle man? They *are* the middle man.
It’s their design, how are they the middle man?
They don’t fabricate the chips themselves, but neither do most silicon designers.
It’s middlemen all the way down. The only person who isn’t a middleman is shoveling sand into a hopper. Or is he merely a middleman for big shovel?
The only difference with the big players is that they are cutting out the resellers and sell in low quantity to the hobbyist.
But in fact you can do it also with most chip makers (as a company). You can even ask for pre programed part if you want.
I hope China guys will flood aliexpress with cheap rp2040 boards in all shapes, colors and flash sizes. Pico is nice but only 2MB and flash chip is not soic-8 so hard to replace. And other boards are too expensive for what they are.
Really? Adafruit and Sparkfun have plenty of board RP2040 options to pick from. Not that expensive either. No need to go to the junk market for your toys.
For my use, the complete ‘board’ package better fills my needs. I would guess that is way it is for most hobbyists. Even at that, Adafruit sells them in singles and 10 packs for the few that have the expertise to design own board around the chip.
Really? Both start with their stuff at more than twice the price of Pico adding stuff I don’t need. I’d prefer more choices like e.g. the WeAct RP2040 board avaliable on aliexpress.
Where does it say they are opening this up to the general public?
terms and conditions on the new site state
`2. Business customers
The Company is a business to business supplier. By ordering, the Customer confirms that it wishes to obtain the Supplies for the purposes of his, her or its business and not as a consumer.`
True, someone didn’t read the big warning right on the front page:
“Raspberry Pi Direct is for direct-to-business sales only.”
The point is that, before this, you needed to go through a private agreement to get the chips in bulk. Now you do not.
A small business is a business. If you’re selling your widgets on Tindie, you’re a business. And if you’re buying lots of 500-3400, you’re very likely a business.
I’m still working through the last of four tubes of ATTiny85s I bought on fire sale back in 2008…
Raspberry Pi keeps upping their commercial game, while the whole educational thing has been going downhill since long before the pandemic. The first menu item on raspberrypi.com is now “For industry”. There is no direct link to the educational resources on the front page or in the menu, just a link to the Foundation at the bottom of the page.
The Raspberry Pi Foundation (raspberrypi.org) seems to have zero resources on the RP2040, one year after launch.
I have no idea how they market half of their products as “for education”, and I don’t know why taxpayers in the UK should pay for this.
.com:
They came out with an entire book about starting out with the Pi Pico and micropython, which is aimed at being more accessible than C or C++ MCU coding. The Pi Pico features fairly heavily on the website (2nd on products), and the book is one of the first things you see when you click on it.
.org:
Foundations are always sluggish on the uptake. In any case, I don’t see any of the commercial arm’s products featured heavily, outside of AstroPi. They look to be emphasizing education and accessability much more, which an SBC is much more useful for anyways. Not sure how them coming out with an MCU invalidates the success of the foundation in their actual goals…
AFAIK, Raspberry Pi Ltd (.com) are a wholly owned subsidiary of the Foundation (.org) which means profits go entirely to the Foundation. The Foundation is a device agnostic educational charity with the advantages that a charity therefor has. As a Ltd, the trading side does not benefit from any `taxpayer cash`. I guess if the Foundation doesn’t have any stuff on the Pico that is their lookout, I guess they just take the money from the sales.
Ah. It’s QFN. I was worried it was BGA. QFN I can deal with. BGA, not so much.
Big deal, ESP32 if far better in every way.
Fragmented memory? Did bite me in the behind when i wanted to emulate a beefy Z80 system with 128 KByte of RAM on it. 520 KByte of RAM? Only if you need that in small chunks, otherwise hope you have PSRAM attached to your system. ¬.¬
This is great news for my latest project, which is a rude Belgian-inspired automaton controlled by a cluster of Raspberries.
I will be calling it the
“Mannequin Pis”
If we could upvote comments here, you’d be getting one. Bravo.
I want them to release in a DIP package with onboard memory, but at least for now, Adafruit has a “QT Py” version the size of a quarter. If you want more pins, though, you’ll have to piggyback solder them yourself…