Google has promised us new hardware products for machine learning at the edge, and now it’s finally out. The thing you’re going to take away from this is that Google built a Raspberry Pi with machine learning. This is Google’s Coral, with an Edge TPU platform, a custom-made ASIC that is designed to run machine learning algorithms ‘at the edge’. Here is the link to the board that looks like a Raspberry Pi.
This new hardware was launched ahead of the TensorFlow Dev Summit, revolving around machine learning and ‘AI’ in embedded applications, specifically power- and computationally-limited environments. This is ‘the edge’ in marketing speak, and already we’ve seen a few products designed from the ground up to run ML algorithms and inference in embedded applications. There are RISC-V microcontrollers with machine learning accelerators available now, and Nvidia has been working on this for years. Now Google is throwing their hat into the ring with a custom-designed ASIC that accelerates TensorFlow. It just so happens that the board looks like a Raspberry Pi.
We all know CERN as that cool place where physicists play with massive, superconducting rings to smash atoms and subatomic particles to uncover secrets of matter in the Universe. To achieve this aim, they need to do a ton of research in other areas, such as development of special particle detectors.
While such developments are essential to the core research needs of the Centre, they also lead to spinoff applications for the benefit of society at large. One such outcome has been the Medipix Collaborations – a family of read-out chips for particle imaging and detection that can count single photons, allowing X-rays and gamma rays to be converted to electrical signals. It may not be possible for us hackers to get our hands on these esoteric sensors, but these devices are pretty interesting and deserve a closer look. Medipix sensors work like a camera, detecting and counting each individual particle hitting the pixels when its electronic shutter is open. This enables high-resolution, high-contrast, noise hit free images – making it unique for imaging applications.
Some months back, CERN announced the first 3D color X-ray of a human made possible using the Medipix devices. The result is a high-resolution, 3D, color image of not just living structures like bones, muscular tissues and vessels, but metal objects too like the wrist watch, seen in the accompanying photograph. The Medipix sensors have been in development since the 1990’s and are presently in their 4th “generation”. Each chip consists of a top semiconducting sensor array, made from gallium arsenide or cadmium telluride. The charge collected by each pixel is transported to the CMOS ASIC electronics via “bump bonds”. The integration is vertical, with each sensing pixel connected via the bump bond to an analog section followed by a digital processing layer. Earlier versions were limited, by technology, in their tiling ability for creating larger matrices of multiple sensors. They could be abutted on three sides only, with the fourth being used for on-chip peripheral logic and wire-bond pads that permit electronic read-out. The latest Medipix4 Collaboration, still under some development, eliminates this short coming. Through-silicon-via (TSV) technology provides the possibility of reading the chips through copper-filled holes that bring the signals from the front side of the chip to its rear. All communication with the pixel matrix flows through the rear of the chip – the peripheral logic and control elements are integrated inside the pixel matrix.
The Analog front end consists of a pre-amplifier followed by a window discriminator which has upper and lower threshold levels. The discriminator has four bits for threshold adjustment as well as polarity sensing. This allows the capture window to be precisely set. The rest of the digital electronics – multiplexers, shift registers, shutter and logic control – helps extract the data.
Further development of the Medipix (Tech Brief, PDF) devices led to a separate version called Timepix (Tech Brief, PDF). These new devices, besides being able to count photons, are capable of two additional modes. The first mode records “Time-Over-Threshold”, providing rough analog information about the energy of the photon. It does this by counting clock pulses for the duration when the signal stays above the discrimination levels. The other mode, “Time of Arrival”, measures arrival time of the first particle to impinge on the pixel. The counters record time between a trigger and detection of radiation quanta with energy above the discrimination level, allowing time-of-flight applications in imaging.
Besides medical imaging, the devices have applications in space, material analysis, education and of course, high energy physics. Hopefully, in a few years, hackers will lay their hands on these interesting devices and we can get to know them better. At the moment, the Medipix website has some more details and data sheets if you would like to dig deeper. For an overview on the development of such single photon detectors, check out this presentation from CERN – “Single X-Ray Photon Counting Systems: Existing Systems, Systems Under Development And Future Trends” (PDF).
Fully autonomous cars might never pan out, but in the meantime we’re getting some really cool hardware designed for robotic taxicab prototypes. This is the Livox Mid-40 Lidar, a LIDAR module you can put on your car or drone. The best part? It only costs $600 USD.
The Livox Mid-40 and Mid-100 are two modules released by Livox, and the specs are impressive: the Mid-40 is able to scan 100,000 points per second at a detection range of 90 m with objects of 10% reflectivity. The Mid-40 sensor weighs 710 grams and comes in a package that is only 88 mm x 69 mm x 76 mm. The Mid-100 is basically the guts of three Mid-40 sensors stuffed into a larger enclosure, capable of 300,000 points per second, with a FOV of 98.4° by 38.4°.
The use case for these sensors is autonomous cars, (large) drones, search and rescue, and high-precision mapping. These units are a bit too large for a skateboard-sized DIY Robot Car, but a single Livox Mid-40 sensor, pointed downward on a reasonably sized drone could perform aerial mapping
There is one downside to the Livox Mid sensors — while you can buy them direct from the DJI web site, they’re not in production. These sensors are only, ‘Mass-Production ready’. This might be just Livox testing the market before ramping up production, a thinly-veiled press release, or something else entirely. That said, you can now buy a relatively cheap LIDAR module that’s actually really good.
Last year, Jiangsu Yuheng Co., Ltd introduced a new microcontroller. The CH554 is a microcontroller with an E8051 core with a 24 MHz clock, a little more than 1 kB of RAM, and a bit more than 14 kB split between the code and data Flash. In short, it’s nothing too spectacular, but it makes up for that with peripherals. It’s got SPI and ADCs and PWM, UARTs, and even a few capacitive touch channels. It’s also a USB device, with some chips in the series able to function as a USB host. You can buy this chip for a quarter through the usual retailers.
Normally, this isn’t huge news. The 8051 is the most copied microcontroller on the planet, and there are probably billions produced each year. Cheap parts are only cheap if your time is free; you’ll usually spend ages trying to digest the datasheet and get a toolchain up and running. That’s where this chip is a little different. There are multiple efforts to bring an Open Source toolchain to this chip. And they’re doing it in Windows and Linux. Someone really cares about this chip.
The current best option for an SDK for this chip comes from Blinkinlabs, with a port of the CH554 SDK from Keil to SDCC. There are real, working code examples for this chip using an Open Source toolchain. Sure, it might just blink a LED, but it’s there. If you can blink a LED, you can do just about anything from there. Programming the chip happens over USB with the ‘official’ WCHISPTool (Windows) or LibreCH551 (command line). The end result is a completely Open Source toolchain to program and upload a hex file to a cheap chip.
There are a few more chips in the CH554 series, ranging from the CH551 in an SOP-16 package to the CH559 in an LQFP48 package, with more features available as the chips get bigger. It’s an interesting chip, with some somehow implementing a USB hub, and could be a very cool chip for some low-level USB hacking.
After exploring a few random online shops one day, [David] (thanks for sending this in, by the way) ran across a very interesting chip. It’s a dual-core, RISC-V chip running at 400MHz. There’s 6 MB of SRAM on the CPU, and there’s 2MB for convolutional neural network acceleration. There is, apparently, WiFi on some versions. There are already SDKs available on GitHub, and a bare chip costs a dollar or two. Interested? Log in to Taobao, realize Taobao does pre-orders, and all this can be yours.
This is a preorder — because apparently you can do that as a seller on TaoBao, but the Sipeed M1 K210 is available as a ‘core’ board with 72 pins in a one-inch square package, a version with WiFi, or as a complete development board with an OV2640 camera, 2.4 inch LCD, microphone, and onboard USB. There are videos of this chip running a face detection routine. It found Obama.
Over the years we’ve seen a few RISC-V chips given development boards, and you can buy them right now. The HiFive 1 is an exceptionally powerful microcontroller with processing power that puts it right up against the Teensy (which is built around a Freescale chip), but it’s also fairly expensive. We’re not sure the Arduino Cinque (also RISC-V) ever made it to production, but again, expensive. The idea that a RISC-V microcontroller could be available for just a few dollars is very interesting, it even comes with SDKs and utilities to make the chip useful.
Since Microchip acquired Atmel, the fields of battle have fallen silent. The Crusaders have returned home, or have been driven into the sea. The great microcontroller holy war is over.
As with any acquisition, there is bound to be some crossover between two product lines. Both Atmel’s AVR platform and Microchip’s PICs have their adherents, and now we’re beginning to see some crossover in the weird and wonderful circuitry and design that goes into your favorite microcontroller, whatever that might be. The newest part from Microchip is an ATMega with a feature usually found in PICs. This is a Core Independent Peripheral. What is it? Well, it’s kinda like a CPLD stuck in a chip, and it’s going to be in the new Arduino board.
The ATMega4809 is the latest in a long line of ATMegas, and has the features you would usually expect as the latest 8-bit AVR. It runs at 20MHz, has 48 K of Flash, 6 K of SRAM, and comes in a 48-pin QFN and TQFP packages. So far, everything is what you would expect. What’s the new hotness? It’s a Core Independent Peripheral in the form of Configurable Custom Logic (CCL) that offloads simple tasks to hardware instead of mucking around in software.
So, what can you do with Configurable Custom Logic? There’s an application note for that. The CCL is effectively a look-up table with three inputs. These inputs can be connected to I/O pins, driven from the analog comparator, timer, UART, SPI bus, or driven from internal events. The look-up table can be configured as a three-input logic gate, and the output of the gate heads out to the rest of the microcontroller die. Basically, it’s a tiny bit of programmable glue logic. In the application note, Microchip provided an example of debouncing a switch using the CCL. It’s a simple enough example, and it’ll work, but there are a whole host of opportunities and possibilities here.
Additionally, the ATMega4809, “has been selected to be the on-board microcontroller of a next-generation Arduino board” according to the press release I received. We’re looking forward to that new hardware, and of course a few libraries that make use of this tiny bit of custom programmable logic.
MEMS, or Micro ElectroMechanical Systems, are the enabling technology that brings us smartphones, quadcopters, tire pressure monitors, and a million other devices we take for granted today. At its most basic level, MEMS is simply machining away silicon wafers to make not electronic parts, but electromechanical parts. The microphone in your cell phone isn’t an electret mic you would find in an old brick phone from the 80s — it’s a carefully crafted bit of silicon, packed in epoxy, and hanging off a serial bus.
Despite the incredible success of MEMS technology, there is still something in your smartphone that’s built on 19th-century technology. Loudspeakers haven’t changed ever, and the speaker in your newest iThing is still a coil of wire and some sort of cone.
Now there’s finally a MEMS loudspeaker A company called USound has developed the first loudspeaker that isn’t just a bunch of wire and a magnet. This is a speaker built from a silicon wafer that can be as small as 3 mm square, and as thin as 1 mm. Since these speakers are built on silicon, you can also add an amp right onto the package. This is quite literally a speaker on a chip, and we’d bet that there are already engineers at Samsung looking at stuffing this into a flagship phone.