Raspberry Pi has shown interest in an integrated machine learning system

The Raspberry Pi Foundation released its first Pi Pico microcontroller board last January, costing just $ 4. Based on the foundation's RP2040 SoC, the Pi Pico has already sold 250.000 copies and 750.000 have been ordered.

At the recent tinyML Talk conference, the co-founder of the Raspberry Pi foundation, Eben Upton, gave a glimpse into the future of the platform. With Pi Pico, the foundation has shown its interest in artificial intelligence And the next iterations are expected to bring significant improvements to machine learning.

The slides presented by Eben Upton at the event show that Pi Pico could serve as a building block for designing specialized boards for machine learning (ML).

In fact, Pi Pico is a small and inexpensive card that integrates the RP2040 system on a chip (SoC) designed by the foundation itself.

This SoC integrates a dual-core Arm Cortex-M0 + chip that operates up to 133 MHz, with 264 KB of static random access memory (SRAM) and 2 MB of onboard flash storage. Small in size (21 x 51 mm), the card also includes a USB port with 26 I / O pins.

“I think it is overwhelmingly likely that there is some other piece of silicon like [RP2040] from Raspberry Pi. I think there is a huge opportunity here: due to its need to run efficiently on processors, the tinyML world has pushed a real focus on good enough primitives. The interesting thing about this world for us is that it is a very static world in terms of what primitives look like, so there is a bit of research interest at the moment in what can be built in the form of a better implementation, something that probably has no more arithmetic performance than a processor core, but doesn't have all the scaffolding around it.

The I / O connector present on single board computers is absent on the microcontroller board, what can be inconvenient. Instead, the base offers perforated pads with ragged edges, as if to highlight where this microcontroller could be used the most.

The platform is also sold in reels of 600 units to be integrated into automated assembly lines. The new microcontroller board is programmable in C language. A development kit that integrates with Visual Studio is provided for this purpose.

The Cortex M0 + does not have a floating point number processor. This aspect is handled through the C language programming SDK. A MicroPython port is also available on the card for handling Python language software. At the tinyML Talk conference, speakers noted that more system-powered cards are needed on an RP2040 chip. So companies like Adafruit, Pimoroni, and Sparkfun are releasing their own hardware, many of which have features not found in the Pi Pico.

Upton said at the event that the in-house Application Specific ICs team Raspberry Pi (ASIC) is working on the next iteration.

The presentation of Upton suggests the team seems to focus on light accelerators for ultra-low power machine learning applications. During his talk at Upton, he presented a slide titled "Future Directions." The slide shows three current generation "Pi Silicon" boards, two of which are from board partners, SparkFun's MicroMod RP2040 and Arduino's Nano RP2040 Connect.

The third is from ArduCam, a manufacturer of cameras based on the Raspberry Pi platform. ArduCam is currently working on ArduCam Pico4ML which integrates machine learning, camera, microphone and display functions in a Pico box.

The last point suggests what the future project could be, which could come in the form of light accelerators, perhaps 4 to 8 multiple accumulations (MACs) per clock cycle. In his speech, Upton said that it is "very likely that there is another silicon segment coming from the Raspberry Pi."

Finally, if you are interested in knowing more about this new version, you can consult the details In the following link.


Be the first to comment

Leave a Comment

Your email address will not be published. Required fields are marked with *

*

*

  1. Responsible for the data: Miguel Ángel Gatón
  2. Purpose of the data: Control SPAM, comment management.
  3. Legitimation: Your consent
  4. Communication of the data: The data will not be communicated to third parties except by legal obligation.
  5. Data storage: Database hosted by Occentus Networks (EU)
  6. Rights: At any time you can limit, recover and delete your information.