Blog for my various projects, experiments, and learnings

Building a Bare-Metal ARM GCC Toolchain from Source

The GCC toolchain is a nice way to get started with bare-metal ARM platforms because it is free, well-supported, and something that many C programmers are already familiar with. It is also constantly improving, and while most Linux distributions include a version of the arm-none-eabi GCC toolchain in their default package managers, those versions are sometimes old enough that I run into bugs with things like multilib/multiarch support.

So after trying and failing a few times to build and install this toolchain, I thought I would share what eventually worked for me. If you are building this on a system without much free disk space, this did take about 16GB of disk space when I ran it – but you shouldn’t need to keep more than ~500MB of that. I ended up building it on an ext3-formatted 32GB USB drive, and that seemed to work fine. You will probably run into errors if you try to run this on a FAT-formatted drive like a new USB stick, though, because the build process will run chmod and that will fail on a filesystem that doesn’t support file permissions. You might be able to make it work with an NTFS drive, but I digress.

Anyways, the first step is downloading the code, so let’s get started!

Keeping Up With the Moorses: Learning to Use STM32G0 Chips

Microcontrollers are just like any other kind of semiconductor product. As manufacturers learn from customer feedback and fabrication processes continue to advance, the products get better. One of the most visible metrics for gauging a chip’s general performance – and the basis of Moore’s Law – is how large each transistor is. Usually this is measured in nanometers, and as we enter 2019 the newest chips being made by companies like Samsung and Intel are optimistically billed as 7nm.

The venerable and popular STM32F1 series is more than a decade old now, and it is produced using a 130nm process. But ST’s newer lines of chips like the STM32L4 use a smaller 90nm process. Smaller transistors usually mean that chips can run at lower voltages, be more power-efficient, and run at faster clock speeds. So when ST moved to this smaller process, they introduced two types of new chips: faster ‘mainline’ chips like the F4 and F7 lines which run at about 100-250MHz, and more efficient ‘low-power’ chips like the L0 and L4 lines which have a variety of ‘sleep’ modes and can comfortably run off of 1.8V. They also have an H7 line which uses an even smaller 40nm process and can run at 400MHz.

Now as 2018 fades into history, it looks like ST has decided that it’s time for a fresh line of ‘value-line’ chips, and we can order a shiny new STM32G0 from retailers like Digikey. At the time of writing there aren’t too many options, but it looks like they’re hoping to branch out and there are even some 8-pin variants on the table. I could be misreading things, but these look like a mix between the F0 and L0 lines, with lower power consumption than F0 chips and better performance than the L0 chips. The STM32G071GB that I made a test board with has 128KB of Flash, 36KB of RAM, and a nice set of communication peripherals.

STM32G0 breakout board

Simple STM32G0 breakout board

So what’s the catch? Well, this is still a fairly new chip, so “Just Google It” may not be an effective problem-solving tool. And it looks like ST made a few changes in this new iteration of chips to provide more GPIO pins in smaller packages, so the hardware design will look similar but slightly different from previous STM32 lines. Finally, with a new chip comes new challenges in getting an open-source programming and debugging toolchain working. So with all of that said, let’s learn how to migrate!

“Bare Metal” STM32 Programming (Part 8): Learn to Debug Timing Issues with Neopixels

I haven’t written about STM32 chips in a little while, but I have been learning how to make fun displays and signage using colorful LEDs,  specifically the WS2812B and SK6812 ‘Neopixels’. I talked about the single-wire communication standard that these LEDs use in a post about running them from an FPGA, and I mentioned there that it is a bit more difficult for microcontrollers to run the communication standard. If you don’t believe me, take a look at what the poor folks at Adafruit needed to do to get it working on a 16MHz AVR core. Yikes!

When you send colors, the 1 bits are fairly easy to encode but the 0 bits require that you reliably hold a pin high for just 250-400 nanoseconds. Too short and the LED will think that your 0 bit was a blip of noise, too long and it will think that your 0 is a 1. Using timer peripherals is a reasonable solution, but it requires a faster clock than 16MHz and we won’t be able to use interrupts because it takes about 20-30 clock cycles for the STM32 to jump to an interrupt handler. At 72MHz it takes my code about 300-400 nanoseconds to get to an interrupt handler, and that’s just not fast enough.

There are ways to make it faster, but this is also a good example of how difficult it can be to calculate how long your C code will take to execute ahead of time. Between compiler optimizations and hardware realities like Flash wait-states and pushing/popping functions, the easiest way to tell how long your code takes to run is often to simply run it and check.

Working Neopixel Timing

Pulseview diagram of ‘101’ in Neopixel. I can’t be sure, but I think the ‘0’ pulse might be about 375 nanoseconds long.

Which brings us to the topic of this tutorial – we are going to write a simple program which uses an STM32 timer peripheral to draw colors to ‘Neopixel’ LEDs. Along the way, we will debug the program’s timing using Sigrok and Pulseview with an affordable 8-channel digital logic analyzer. These are available for $10-20 from most online retailers; try Sparkfun, or Amazon/eBay/Aliexpress/etc. I don’t know why Adafruit doesn’t sell these; maybe they don’t want to carry cheap generics in the same category as Salae analyzers. Anyways, go ahead and install Pulseview, brush up a bit on STM32 timers if you need to, and let’s get started!

Learning how to FPGA with ‘Neopixel’ LEDs

Whenever I talk to someone about FPGAs, the conversation seems to follow a familiar routine. It is almost a catechism to say that ‘FPGAs are very interesting niche products that, sadly, rarely make sense in real-world applications’. I often hear that organizations with Money can afford to develop ASICs, while hobbyists are usually better served by today’s affordable and powerful microcontrollers except in some very specific circumstances like emulating old CPU architectures. I don’t have enough experience to know how accurate this is, but I do have a couple of projects that seem like they could benefit from an FPGA, so I decided to bite the bullet and learn the basics of how to use one.

I chose a popular $25 development board called the ‘Icestick‘ to start with. It uses one of Lattice’s iCE40 chips, which is nice because there is an open-source toolchain called Icestorm available for building Verilog or VHDL code into an iCE40 bitstream. Most FPGA vendors (including Lattice) don’t provide a toolchain that you can build from source, but thanks to the hard work of Clifford Wolf and the other Icestorm contributors, I can’t use “maddeningly proprietary tools” as a reason not to learn about this anymore.

One thing that FPGAs can do much better than microcontrollers is running a lot of similar state machines in parallel. I’d eventually like to make a ‘video wall’ project using individually-addressable LEDs, but the common ‘Neopixel’ variants share a maximum data rate of about 800kbps. That’s probably too slow to send video to a display one pixel at a time, but it might be fast enough to send a few hundred ‘blocks’ of pixel data in parallel. As a small step towards that goal, I decided to try lighting up a single strip of WS2812B or SK6812 LEDs using Verilog. Here, I will try to describe what I learned.

Icestick with lights

Blue Icestick

And while this post will walk through a working design, I’m sorry that it will not be a great tutorial on writing Verilog or VHDL; I will try to gloss over what I don’t understand, so I would encourage you to read a more comprehensive tutorial on the subject like Al Williams’ series of Verilog and Icestorm tutorials on Hackaday. Sorry about that, but I’m still learning and I don’t want to present misleading information. This tutorial’s code is available on Github as usual, but caveat emptor.

Basic MSP430 Hardware Design

Evaluation boards are great, but eventually you’ll want to make a design which needs to fit in a smaller space, or which uses a type of chip that doesn’t have a cheap board available. When that happens, you’ll often want to design a PCB. And it seems like most microcontrollers have similar basic hardware requirements; decoupling capacitors on the power pins, maybe a pull-up resistor and filtering capacitor on the reset pin, and a few pins which are used for debugging and programming. The MSP430 is no different, although it does have a few small quirks to be aware of.

And while I haven’t found a cheap dedicated USB device for programming MSP430 chips, you can use the debuggers built into TI’s Launchpad boards to program and debug a custom board using their “Spy-Bi-Wire” protocol. So in this tutorial, I’ll go over a basic circuit design for an MSP430FR2111 chip. It comes in a TSSOP-16 package with 3.75KB of FRAM and no Flash memory.

Simple MSP430FR2111 breakout boards

A simple example MSP430FR2111 breakout board design.

I’ll also go over the differences between programming a ‘pulsing LED’ example for the MSP430FR2111 and the MSP430G2553 that we used in the last two examples, as well as how to connect a Launchpad board’s debugger to upload programs to the custom board. So let’s get started!

Basic Low-Power Design: Sleepy PWM on the MSP430

Recently, I wrote about setting up a basic ‘hello, world’ program for an MSP430 microcontroller. These chips look like they are designed to make low-power designs easy to code, so it seems like a good idea to start learning about their low-power sleep modes. TI claims that, at 3V and a 1MHz core clock speed, an MSP430x2xx will consume about 300μA while running, 55μA in “low-power mode 0”, and as little as 0.1μA in “low-power mode 4”. But you may have difficulty reaching those figures on a Launchpad board because of how the circuit is laid out.

You can see section 2.3 of the reference manual for more details on these sleep modes, but the most important thing to know is that different low-power modes can selectively disable the core CPU, system clocks, and peripherals on the chip. Some clocks and peripherals remain active in some low-power modes. For example, the chip’s timers can continue to run while the CPU is off as long as the system clock they are connected to is still active.

So in this tutorial, we’ll walk through pulsing a common-cathode RGB LED through various colors using the MSP430’s timers to generate PWM signals while the chip rests in LPM0 sleep mode. We need to set up the timers and attach an interrupt to periodically change the PWM duty cycles, but once that is done we can turn off the CPU. The timer interrupt will periodically wake it up, and the chip will automatically go back to sleep after the interrupt handler returns.

RGB Common-Cathode LED

“Common-Cathode” LEDs let you control a red, green, and blue LED individually using PWM signals. Each LED is connected to the same cathode (‘ground’).

This post’s Github repository contains two projects; one which dims and brightens a single on-board LED using PWM, and one which uses a timer interrupt to pulse each color in a common-cathode RGB LED while putting the device to sleep when it is not active.

“Bare Metal” MSP430 Programming: Learning About a New Microcontroller

Most of the embedded programming that I’ve written about so far has focused on the STM32 family of ARM microcontrollers produced by ST Microelectronics. Those are a reasonable starting point for learning about microcontrollers, because they have a solid suite of open-source development tools, a decade-long history of popular use, and affordable hardware development tools which let you get started for less than $20.

But now that platforms like Arduino have become popular and presumably profitable, more vendors are getting on board with providing affordable “DIY” development tools. So while I couldn’t find any cheap and widely-available USB debuggers for TI’s MSP430 core, the “Launchpad” boards that they sell each include an on-board debugger which they say can talk to most MSP430 chips through a two-wire interface similar to the “Serial-Wire Debug” protocol used by a number of Mobile ARM cores including the STM32.

MSP430G Launchpad

Like the Arduino Uno, some Launchpad boards use DIP chips which you can remove and plug into a breadboard once they have been programmed.

So let’s learn how to write bare-metal programs for the MSP430! This will be easier than bootstrapping a build system for the STM32, because TI opted to design a special-purpose 16-bit RISC core for these chips instead of buying a cookie-cutter core from ARM. That means that the MSP430-GCC toolchain can include ‘glue’ code like linker scripts and vector tables for every MSP430 chip ever made, and we will not need to write our own. There are also fewer steps involved in setting up clocks and peripherals, so this is a comparatively easy platform to develop on. And as usual, this tutorial’s code is on Github.

Making a Retro Wall Clock

It’s good to know what time it is, and once you work out how to use a RealTime Clock module, making a clock display seems like a natural next step. 7-segment LED displays used to be the go-to way to display time digitally before LCD and OLED screens got so cheap, and I thought it would be fun to make a more modern take on them with diffused multicolor LEDs for the actual lighting. I would like to write a post about drawing to traditional 7-segment digits using shift registers, but this project uses the easy-to-wire ‘Neopixels’ that we all know and love.


The one on the bottom doesn’t know about daylight savings, but you get the idea.

You’ll need a few materials for this project if you want to follow along:

  • Access to a laser cutter and a 3D printer.
  • About 6 x 12″ of 1/8″-thick ‘frosted’ acrylic, to diffuse the LEDs. (About 150 x 300mm, 3mm thick)
  • Enough plywood to make a clock face for the front and back; I wound up using about 500 x 200mm each.
  • 58 individually-addressable RGB LEDs. (28 sets of 2 for each segment of the 4 digits, and 2 more for the colon dots.)
  • A DS3231 or similar realtime clock module.
  • A microcontroller to run everything; I used an ATMega328 on an “Arduino” nano board to keep the code simple.
  • Craft supplies: solder, soldering iron, glue, hot glue gun, wire, etc.

For the LEDs, I cut lengths off of a reel of LED strip that I ordered off Amazon; I think it used SK6812 LEDs and had 30 LEDs / meter. For the two dots of the colon, you can also buy small single-LED boards that are about 10mm in diameter and usually have WS2812B LEDs. I didn’t run into any problems stringing SK6812 and WS2812B LEDs together with Adafruit’s “Neopixel” library, but caveat emptor.

Replacing Broken Car Speakers

Some Cars are Old

Recently, I bought a very used car in the hopes of modifying it to use a parallel-hybrid drivetrain. I’d like to cut the driveshaft and slot an electric motor inline with it, and that will probably be the subject of future posts if I can find a garage willing to do most of the heavy lifting and cutting.

But in the meantime, what sort of car do you get if you’re thinking of cutting its driveshaft in half? A cheap one, of course – and my choices were further limited by a few basic requirements:

  • The car should actually be a small pickup truck; the bed is an ideal place to put the heavy batteries, they tend to have a lot of ground clearance for fitting a bulky motor underneath the chassis which means little or no changes to the bodywork, and passenger space is not a consideration in this build.
  • It should be rear-wheel drive. Since I am planning to add a motor between the transmission and the differential, there needs to be enough space along the driveshaft to fit the motor. That means either rear-wheel drive with the engine in the front, or front-wheel drive with the engine in the back. And I don’t know of any rear-engine pickup trucks.
  • It should have an automatic transmission. This kind of simple conversion leaves you with two acceleration pedals, and I don’t have enough feet to use four pedals at once.

So I wound up with a cheap-and-cheerful 1994 Toyota Pickup Truck. It doesn’t actually have a model name; in North America, the Hilux brand was renamed to “Truck” in the 1970s before eventually being replaced with the Tacoma. So this is not a luxury vehicle. It has roll-up windows, it does not have ABS, it came with a broken speedometer, and more importantly its speakers had disintegrated in the decades since 1994.

Old Speaker


That’s no good, and it sounded like the buzzing of a thousand wasps. Plus, replacing your speakers seems like a pretty easy first project if you want to learn about a car. You only have to remove a few interior panels, and your sound system is a safe thing to break compared to say, your brakes. So let’s get started! These specific instructions may be for a 1994 Toyota pickup, but I’ll bet the process looks similar for the front speakers of other cars.

Recycling HDPE: What Doesn’t Work

Over the past month or two, I’ve been learning about plastic recycling with a friend at a local makerspace. We decided to start out with HDPE as a nice beginner’s material, because it melts at less than 200C and tends to be fairly nontoxic. We’ve tried a few different techniques now, and we’ve learned a lot about what doesn’t work.

This is an exciting field and we hope to make a decent press for forming 3mm and 6mm-thick HDPE sheets soon, but our early designs have shown us some key difficulties to be aware of if you are planning to start recycling plastics on your own. And I’ll mention this later, but please research what you plan to recycle ahead of time and avoid plastics which might release hazardous fumes. As you will see, I am not an expert.

Our first instinct was to imitate an injection-molding press, and we started by scavenging some large heating elements from a Goodwill-sourced griddle. It consisted of four ~13-Ohm coils arranged as two parallel banks, each consisting of two elements in series. That meant an effective resistance of 13 Ohms, or a bit over 9 amps at the mains 120V which it was more or less directly connected to.

We wound up clipping the heating coils out and scavenging some of the crimped and heat-insulated wiring, strapping them to a smooth-bore metal pipe, and then soldering or spot-welding the four coils back together in their original 2×2 arrangement. Temperature monitoring is all well and good, but this was a first attempt; we just wired the coils across a 2-prong plug, and flipped a surge suppressor on and off to control the heat. A handheld infrared thermometer gave us a rough indication of the tube’s temperature, and we aimed for 350-400F to melt the HDPE.

Our first attempt at a plastic-melting tube. It's probably mostly safe.

Our first attempt at a plastic-melting tube; the heating elements are the thick copper-coated wires, and the beige wires connect directly to a 2-prong wall plug.

“Bare Metal” STM32 Programming (Part 7): Embedded C++ Inheritance

As you start to re-use components like sensors and displays, you might start to get frustrated with how long it takes to set up new projects. Copying and cleaning code between old working examples and new ideas can be time-consuming and tedious. It’s much easier to simply copy a few portable library files around. While there are plenty of existing libraries for these sorts of peripherals and external devices, it’s good to learn how to write your own, and this is also a good way to demonstrate a few ‘gotchas’ that you should be aware of when using C++ in an embedded application.

In this tutorial, we will walk through setting up a couple of object-oriented classes to represent a common communication ‘I/O’ peripheral model:

One example of what a simple common communication model could look like.

One example of what a simple common communication model could look like.

For simplicity’s sake, I’ll only cover a class for a bank of GPIO pins to demonstrate the core requirements for using C++ in an embedded application, but you can also find similar classes for the I2C peripheral and an SSD1306 OLED display in the example Github repository’s reference implementation of the concepts presented in this tutorial.

“Bare Metal” STM32 Programming (Part 6): Multitasking With FreeRTOS

In a previous tutorial, we walked through the process of setting up a hardware interrupt to run a function when a button was pressed or a dial was turned. Most chips have a broad range of hardware interrupts, including ones associated with communication and timer peripherals. There is also a basic ‘SysTick’ timer included in most ARM Cortex-M cores for providing a consistent timing without needing to count CPU cycles.

One good use of that evenly-spaced timing is to run a Real-Time Operating System (often called an ‘RTOS’) to process several tasks in parallel. As you add more parts to your projects, it will become awkward to communicate with them all using a simple ‘while’ loop. And while hardware interrupts can help, it’s usually good to do as little as possible inside of an interrupt handler function.

So let’s use FreeRTOS, an MIT-licensed RTOS, to run a couple of tasks at the same time. To demonstrate the concept, we’ll run two ‘toggle LED’ tasks with different delay timings to blink an LED in an irregular pattern.

Example LED Timing

Example LED timing with two ‘toggle’ tasks delaying for different amounts of time.

This example will also add support for some faster microcontrollers; the STM32F103C8 and STM32F303K8, which are Cortex-M3 and -M4F cores respectively. The F103C8 is available on cheap ‘blue pill‘ and ‘black pill‘ boards, and ST sells a ‘Nucleo’ board with the F303K8. Both of those chips can run at up to 72MHz, and they can also get more done per cycle since they have larger instruction sets.  And as usual, there is example code demonstrating these concepts on Github.

The W1209: A (sometimes) STM8-based digital thermostat

Among the cheap gadgetry that constantly spews forth from the spawning pits of consumer electronics, sometimes you can find a gem. The W1209 is an interesting board which is designed to act as a thermostat which can also switch a relay when certain temperatures are reached. It is used in everything from rice cookers to yogurt machines to people who want their A/C to only turn on when it’s hot out. Originally they shipped with a fairly powerful STM8S003F3 microcontroller, and they cost less than $1.50 each.

Even so, their popularity has inspired the usual imperfect cloning process, so chances are that a board you buy from ebay, Aliexpress, Taobao, or Amazon today will need some touching up before you will be able to reprogram it. In this tutorial, we will replace the microcontroller and add a few missing capacitor/resistors. Then we’ll upload a simple test program to test that the microcontroller and board both work. So here is an example of what is probably the worst possible knock-off that you could fear to get, with missing or faulty parts highlighted:

Problematic W1209

W1209 that needs fixing

Well, that’s life. You try your best, and then someone comes along and kicks you in the teeth. But let’s make some lemonade and take the opportunity to learn about fixing cheap-o knock-off circuit boards!