Most important thing is to have a passion project that you are extremely motivated to complete. I'm always baffled when folks try to learn an extremely deep domain by just watching random videos and following tutorials.
ChatGPT is an incredible engineering mentor. You can copy/paste chunks of your KiCAD schematic into the chat and it will just jump in and start recommending parts that meet the moment. Just ask it everything, and keep asking follow-up questions.
Make no mistake; you aren't just learning embedded development... you're learning electronics. That means you should get into the mindset that you'll be doing a lot of skill stacking.
If you don't know how to CAD things in Fusion and 3D print them, fix that. I'm happy to recommend an excellent tutorial video.
I recommend getting a bag of 25 of each of the ~100 most common resistor values and putting them in a bin. This will cost ~$100 and you will have all of the resistors. I cannot tell you how great this is. It's like having a jar of every spice.
Amazon has many vendors that sell plastic snap cases of capacitor values and various diodes. I have a stack of different parts. Again, all of the spices.
IMO the most underrated microcontroller is the ESP32-P4. It's a powerhouse, and a company called Wireless Tag makes a module called WT0132P4-A1. It is the bee's knees. RP2354A is also an amazing thing because it has something called PIO that will make you very happy when the time comes.
Not all breadboards are created equal. Many are terrible. Invest in BB1660.
KiCAD has become really, really good. And JLCPCB is both cheap and flexible.
If you get curious about surface mount - and you will - if you remember one thing from all this, it's to invest a few hundred in a decent STEREO microscope on AliExpress. If you give your brain the ability to perceive depth, you'll suddenly realize that you're a soldering ninja. (A Robin's Tools Pixel Pump doesn't hurt, either.)
rkagerer 3 hours ago [-]
Way too much caution in this thread. I've done all kinds of programming including embedded and full stack (from before the latter was even a colloquialized term).
I absolutely love the degree of control you can get at the microcontroller level. One of my proudest projects had IRQ handlers written in assembly where I counted instruction cycles in order to guarantee worst-case latencies and well-defined impact to realtime activities in the main loop. There were also some lower-priority tasks going on that were cooperatively multitasked.
Some time later I wrote a JavaScript library that mimicked some of those cooperative multitasking concepts to queue background image downloads and other generic "ITasks" in a manner that maximally utilized available connections (which were extremely limited by servers back then) and allowed on-the-fly reordering. The frontend used it to always ensure whatever image the user was most interested in (i.e. tiny previews, and large version of the result presently hovered over or that we anticipate will be hovered next) came up as close to instantaneous as we could get, and the rest of the large images lazy-loaded in an intelligent order. The result was an experience that felt so much faster than how most other websites operated back then (and ironically even now as bloat has really slowed a lot of sites down), and users loved it.
So the concepts you learn at one end of the spectrum can definitely benefit you at the other.
I have much less fun now that so much of the web has moved to React and other cookie cutter frameworks where the level of control comes at a higher cost (in terms of learning curve, understanding all the layers of abstraction, and space for developer creativity).
In other words, just get in there and start playing with a personal project or two, and trust your instincts and all the experience that honed them!
ChrisMarshallNY 7 minutes ago [-]
That's a pretty big transition.
Get comfortable with breadboards, scruffy command-line interfaces, and hex dumps. You'll need to stare at a block of raw memory, and determine the data structure, therein.
I started with embedded. It's a lot of fun, but you'll spend a hell of a lot of time, preparing the stage. You will be responsible for all the low-level stuff that a modern host OS does.
I'd actually suggest starting with pure Machine Code, first. It will make ASM look easy, and help you to really understand the processes in the chips.
Also: IRQ is The Way. It's a whole new way of looking at threading. There are absolutely no safety nets.
That said, it's been 35 years, since I did that low level work. I suspect the tools have come a long way, since.
snailcoder 11 minutes ago [-]
I think if you have a genuine interest in solving problems where that tooling is the best solution, it should be an easy transition. Especially being able to use generative AI as a resource.
Now, if you’re just over, focused on technologies and want to use a technology for the sake of using that technology, probably won’t be as easy.
Saying you want to transition to C, C++, or Rust is like a handyman saying they want to to get better with electrical tools vs plumbing tools or circular saws vs reciprocating saws. They focus on the problem than try to pick the right tool for the job.
Focus on what you want to solve, and then apply the tooling. Solve the same
Problems you’ve already been solving, with your new tooling and weigh the pros and cons of both. The obvious choices will come out based on the problem at hand. The jump from web apps to firmware may be a big jump. But if you focus on a problem that interests you, and it appears those techs may be a better tooling for your task. Maybe solve it in your traditional languages first, then re apply the new stack. Or contribute to tools you’re already using, written in those stacks.
ferguess_k 3 days ago [-]
I'd say just do it. You have been in the industry for 15+ years and you surely know how to network/do interviews. I feel that you are just hesitating or wait for some confirmations. I wrote similar posts so that's my guess, but please criticize me if I'm wrong.
My point is, whoever has actually done the transition, their experience may not really help you a lot. They might have a cousin who actually refers them to the company; They might have worked 10 years of embedded projects as a hobby; They might be lucky because 5 years ago there were a huge opening somewhere. I heard of a metaphor about elevators -- Several people ride an elevator to floor 100, and a reporter asked them how they get there -- one of them said he did 1,000 push-ups, another said he read 2 books, and a third said he bumped his head against the elevator's walls.
Those experience might be very well relevant, or not. But above all, copying others' experience, in THIS market, probably doesn't work. Now that you are 6 years senior than me regarding SWE experience, I think you already know what to do -- just do it.
egeozcan 3 hours ago [-]
> I'd say just do it.
I clicked on the title to write a comment saying exactly this. I didn't know anything about the experience or skills. Unless you're trying to get into Enterprise Java without knowing what a bean is, just doing it usually works.
jraph 1 hours ago [-]
> Unless you're trying to get into Enterprise Java without knowing what a bean is
Can't you just learn what a bean is on the spot as well?
ipnon 4 hours ago [-]
I agree Nike style is the right move. The caveat is that we used to have a liquid and legible labor market for pricing and allocating computer talent. This is in flux now. I think if you want to keep working in computer technology you need some more chutzpah than was required a decade ago. “Grind Leetcode and spam job boards” won’t get you far these days. It’s better to network, make durable relationships, focus on solving persistent and novel problems. We have to be real engineers once more instead lavishly remunerated corporate monkeys.
tptacek 2 hours ago [-]
I started my career in C/C++ and have done a fair amount of embedded work. In terms of actually doing stuff, there has never in the history of the world been a better time to do low-level and embedded systems work. The right advice here: find a project, scrounge the parts, and have an agent do basic bring-up work for you. If your last serious C work was 15 years ago, I think you'll be astonished how quickly you can progress to the "rewarding" part of any project.
If your question is instead how you make a career transition to embedded/firmware work, that's much trickier. The obvious concern is AI's impact on the job market, but I'm chiming in to observe that this isn't the biggest problem with a web->embedded transition. Good jobs doing embedded stuff, at least for people who don't already have a pedigree doing that work, has for as long as I can remember been a bottleneck for people wanting to do this stuff. There's simply a lot more high-level work to be done than low-level work.
I can see bright spots potentially ahead; I'm not a big Jevon's Paradox person but one kind of clearly Jevon-y area is hardware, because we've zorched one of the biggest impediments to getting new projects off the ground. But it hasn't happened yet, and might not.
throwup238 19 minutes ago [-]
> In terms of actually doing stuff, there has never in the history of the world been a better time to do low-level and embedded systems work.
Couldn’t agree more. I just had to redesign a PCB from an old client and it took me less than a tenth of the time and money that it took me to make the original fifteen years ago. What used to be $30,000 for ten assembled boards is now $2,000 shipped. Finding all the footprints and reference designs was faster than ever, what took two months end to end took a week (not including shipping wait times).
HALs and vendor libraries are in a far better state so writing the firmware felt less like slogging through a MCU reference manual and more like actual programming. LLM to do the brute forcing so that peripherals like the DMA channels play nicely together, which always burned up so much time during bringup.
Hell I was tempted to just through ESPHome on the firmware and use one of their modules with some custom C. I was dreading the redesign because I hadn’t designed any PCBs in a while but I had an absolute blast.
Just start with some cheap ESP32 boards from Amazon, a breadboard, and a wire terminal breakout for the ESP32
coryrc 2 hours ago [-]
> Good jobs doing embedded stuff, at least for people who don't already have a pedigree doing that work, has for as long as I can remember been a bottleneck for people wanting to do this stuff.
QFT
--former(?) embedded engineer who moved into devops
AnimalMuppet 1 hours ago [-]
> There's simply a lot more high-level work to be done than low-level work.
Absolutely true. I'd say it's different by a factor of 10.
But on the flip side, there are a lot fewer people with the skills and/or willingness to do embedded. So it works out to pretty steady employment, at least in my career. Be aware, though, that the smaller market can mean more time spent looking before you can find a job.
yard2010 53 minutes ago [-]
When I went to uni, they had this fun iot course that a company from the area ran as a hiring campaign. Ill brag shamelessly: after the course was over they tried to hire me and I politely declined to keep working in web dev. So maybe find how can you take an iot course in the nearby uni, it might be even free, I feel like this industry is small (at least where I come from) so you can make nice connections, and even if not, well at least you end up with a nice project to show.
By the way, this course was hands down the most hardcore thing I ever did. At one point I had to count high pulses for a number of c for loop ticks to read a temperature, in the process of writing the driver by hand (no AI yet), ending up with too high readings, and I couldn't debug it since it's actual pulses of electrons in the wire, once you signal it to start you can't stop it. In the end it was an off by one error, but not the usual kind, I shifted left 1 bit too much. I got stuck like 2 hours in the library trying to figure out what went wrong (I always suspect it's the hw fault, naturally), so that aha moment was really fun!
There is something special about creating something physically. I wish you make it soon and have fun!
brudgers 39 minutes ago [-]
I see two layers here. One is doing embedded development. The other is getting someone to pay you for it.
Embedded development has never had a lower bar to entry. There are cheap development boards, free tools, and modestly priced test equipment. Doing embedded development is entirely up to you.
Getting someone to pay for your time doing it is just the generic problem of getting someone to pay for your time doing something you would prefer to do. It depends on other people and your social capital
Who you know, how good a hang you are [1], the economics of the field, your opportunity cost and compensation requirements, etc. are what matter most. Luck aside there’s no free lunch. The place to start is your building your network. That’s hard.
[1] technical skill might or might not be a determining factor within a workplace culture.
unrented7977 4 hours ago [-]
If not embedded, there is still plenty of space for low level PC programming in C/Rust/etc. Try as we might, I don't think we will ever get rid of C as the native language of low level Win/Linux APIs. Someone has to write low level drivers.
But embedded is something else entirely that I'd argue is fundamentally incompatible with the perspective of a web dev. In embedded, you need to be aware at every moment of exactly what the CPU is doing and how many resources are being used. You generally don't have multiple CPUs or even threads. Multitasking means hardware interrupts instead of cooperative threads. You also likely need to be aware of assembly because sometimes you have to write a really hot function in raw assembly to get the required performance.
Embedded is all about resource optimization, and we dev is about throwing more and more resources at the problem. It will probably be a difficult transition to make because it's a fundamentally different way of looking at and thinking about programming.
I would recommend you find a problem, then buy the cheapest microcontroller you can find and spend a few weeks learning how to optimize your code to make maximal use of the hardware while still solving the problem in acceptable time. A big part of getting my current job was talking in my interview about how I solved a novel problem by using an obscure CPU feature in assembly that resulted in 30x fewer instructions than optimized C.
Find a problem, solve the hell out of it, and speak confidently about your solution.
One benefit that web dev does bring you is thinking about concurrency. Your CPU can only do one thing at a time, but microcontrollers have peripherals that run in parallel, and the embedded system usually consists of multiple chips talking to each other and doing independent work. Managing this concurrency efficiently with a single CPU thread is a large chunk of embedded engineering. Instead of IPC, you do things like set up a DMA pipe between SPI and RAM, trigger an interrupt on I2C bytes, use harhardware timers to simulate threads. You'd be shocked at just how much concurrent work can happen with only a single non-pipelined in-order CPU thread.
If you want to break into a new field, the advice is always the same: do the work in your own time, build a portfolio, show you know the job, and speak confidently about the hard problems you learn to solve.
zergrush 2 hours ago [-]
One problem I found while reading your argument is that it completely ignores the reality where LLMs are doing exactly what you described without the same level of human effort.
The "hard" stuff you described is already being done with LLMs
kooi 15 minutes ago [-]
Check out Rust with RTIC or Embassy on ESP devices. The tool chain is fairly well developed and you'll get experience with Rust and microcontrollers.
wernam 2 hours ago [-]
For fun or for business?
For fun - just do it. There are no barriers. Esp32, audio synthesizer gadget, cameras. Everything is cheap and available.
For the money? Expect lower pay, lower standards, worse hours and shit conditions
coryrc 2 hours ago [-]
That's why I moved away from it (money).
AnimalMuppet 4 days ago [-]
There are two issues: 1) transitioning your skills, and 2) persuading someone to hire/pay you for your new skills.
Transitioning your skills is possible. But embedded is a different world and a different mindset. You a lot less RAM and a smaller CPU. You may have hard deadlines - the right answer can be wrong if it's too late. You may have multiple threads that have to share data and not corrupt it, no matter what order events occur in.
In the worst case, you may not have logging available at all, and you may not have a debugger. (I think the worst I ever had to do was attach a logic analyzer to the address bus, triggered by a specific address, and then use it to look at past memory addresses in order to figure out what path of execution had got me to the point of the crash.)
So, yeah. It's a different world. Learnable, but different.
Getting someone to believe you can do it enough that they're willing to pay you: I don't have great advice there. I spent almost my whole career in embedded. It seems to me that it's easier to get from there into, say, desktop programming, than it is to go the other way. Your best bet might be to emphasize your native development background, and count on there being a shortage of embedded people.
wojciii 13 hours ago [-]
If you have no logging and no debugger .. you should run away. Fuck debugging something complex by blinking a led. This is a waste of your time.
I'm sure you can make Claude help you with this .. but still..logs are important. So is a good debugger and software support for it. It saves so much time in the end.
aforwardslash 3 hours ago [-]
Having done basically that a couple of weeks ago (with a scope to verify timing signals), its actually pretty common, specially when you are implementing protocols with hard timing requirements.
And in my case, I was actually troubleshooting the code that makes leds blink, with a bit-banged implementation on a deterministic processor.
voakbasda 3 hours ago [-]
Have you never experienced a bug that goes away from adding/removing a logging statement?
I recently restructure a logging module in a client’s project, and the resulting performance gains exposed all sort of race conditions.
Sometime all you get is an LED.
studog 1 hours ago [-]
I had a bug triggered by adding a comment. :-(
unrented7977 4 hours ago [-]
> This is a waste of your time.
This is the reality of highly constrained embedded systems. If you don't like that, I'd suggest moving to web dev where you have state of the art debugging.
readthenotes1 4 hours ago [-]
That's kind of like saying "don't be a farmer--you have to go outside".
mikewarot 3 days ago [-]
You can get started learning with any of the amazingly cheap and powerful ESP32, Raspberry Pi, Raspberry Pi Pico, and Arduino stuff out there, along with dirt cheap TANG FPGAs, etc. Get some projects under your belt in your free time. Start with the canonical blinking of an LED, and step it up from there.
jon-wood 4 hours ago [-]
It’s totally doable, as with all things in this industry the bulk of the work is proving you can already do it before someone employs you. Build some toy projects, learn how to use embedded Rust or C, maybe write some blog posts but at the very least have some code on GitHub you can point potential employers at.
You’re also going to have to get comfortable with some level of electronics, because it’s very rare someone wants embedded development done that doesn’t interact with the outside world in some way, and that’s going to require knowing how to drive some IO.
rleigh 1 hours ago [-]
I transitioned from C++ native development at the age of 39, 7 years ago now. Haven't ever looked back; embedded development is a lot of fun and I'm constantly doing new stuff, be it exploring a new peripheral and making it work just so, or integrating lots of different activities into an RTOS and making sure the timings are still spot on. Yesterday was implementing a radix tree. Or doing mechanical design and test fixtures. There's a lot of variety, and the hands-on application of software, electronic and mechanical aspects to make working systems is very satisfying.
Not having an electronics background or an embedded background, there was certainly a steep learning curve. Lots of stuff to learn and master. But anyone who already knows C and C++, and who is prepared to knuckle down and get stuck in, will do fine.
dehrmann 25 minutes ago [-]
Not advice for switching, but I've worked on both, and one interesting commonality is special cases for bug handling and vendor compatibility.
qazxcvbnmlp 3 hours ago [-]
As someone who currently does embedded, and transitioned into it with extremely little professional experience. You need to be able to demonstrate that you understand the tradeoffs that come with embedded and how you can provide value to someone who would hire you. Web development is very much a continuous integration easy to fix bugs kind of environment, in embedded bugs can have real cost.
As someone who frequently hopped around across different domains, the key is demonstrating that you understand their problems and can provide value. Also, being curious helps.
Also, today with AI am not sure how much native development is going to matter. A handful of best practices in an already set up project plus an AI agent is gonna spit out very workable code. Right now the problem space is very much about reasoning about the larger system, de bugging stuff, thinking about how pieces work together, going in the lab and connecting debuggers so that they agent can do stuff to the hardware, etc.
zergrush 2 hours ago [-]
> Right now the problem space is very much about reasoning about the larger system, de bugging stuff, thinking about how pieces work together
it's already being done without humans, you are already far too behind here
hingler36 4 hours ago [-]
IMO the single most important thing about embedded that doesn't exist in web dev is timing assurances. Juggling several competing tasks and making sure each one gets completed at exactly the right time is not an uncommon constraint in this space. My suggestion would be to pick a toy example like motor control or something and use it to learn about state machines.
xesni 11 minutes ago [-]
people here say "just do it", but i don't think that it is that simple - you have to have some background/projects etc. to even get an interview...
mkozlows 24 minutes ago [-]
Instead of saying "hey claude, I need you to build this web page for me" you say "hey claude I need you to build this firmware for me."
don't know why everyone's acting like this is hard, seems straightforward
rozenmd 3 hours ago [-]
Just do it.
I started with data engineering around 2015ish purely because python was easy, and moved into frontend because I needed to make the data understandable for execs. Start building projects and showing them to people, get feedback and iterate. Nothing has changed.
bruce511 4 days ago [-]
I've done both over my career (as well as some other stuff.)
While they are very different, if you know C, it's not hard to transition.
Getting a job in the embedded space though is obviously a lot harder than web. There are simply fewer jobs to be had. Doing some personal projects in the space may reawaken your C skills, and help demonstrate competence.
Good luck!
svnt 3 hours ago [-]
I’d ask why you want to change and what kind of embedded work? Demand is lower and more specialized. Embedded is almost a solved problem by AI in many cases. The compiler plus physics proximity to ground truth means you give the agent the tools a debugger and output access and it does a more thorough job than maybe 90% of embedded devs.
If part of the reasoning is you are hoping to escape implications of AI I don’t think this will help.
gortok 3 hours ago [-]
I’m not sure this is accurate, at least it doesn’t match my experience.
Firmware, especially for the various SoCs, is not a solved problem. There’s far fewer examples of publicly available code for LLMs to learn from, and documentation is at times behind paywalls.
While I’m sure on any long enough time horizon this will change, at the moment finding useful code for some of these SoCs is a lesson in pain. I would also expect relying on an LLM for correct code for low level work like this would be a mistake at best. There are subtle bugs that you can’t shake out with code review and it takes far longer to debug embedded devices than it does web systems.
mirmor23 9 hours ago [-]
> I'd like to transition closer to my roots with a job doing primarily c, c++ or rust. The natural place seems to be in embedded development or firmware work. How feasible is the transition this late in the game? Any advice?
Have you done any self-directed project to further this ambition? the answer depends on your current state..
Theodores 3 hours ago [-]
Try the 'Brian Eno' approach. When the early synthesizers came out, he was one of the few to read the manual for the Yamaha DX-7. Everyone else was getting the same sounds out of this synth, except for Eno. They had one-hit wonders, meanwhile, Eno's album Apollo was the start of a very long career, where RTFM was crucial to his success.
Much tech comes along, for example, I think USB-C is brilliant, and I am sure that, if I knew how to make USB-C gadgets, there would be a job there for me, albeit in China, because we don't make anything in the country I am in.
I can use USB-C now, no brain cells required, but nobody is going to give me a job because I know how to plug my phone in. Yet, if I knew the USB-C specs because RTFM, then it would be a whole different ball game, even amongst developers, where they might be adapting reference implementations rather than taking a RTFM + first principles approach.
coryrc 1 hours ago [-]
In Bluetooth, that's because nobody follows the specs because you'd break a bunch of nonconforming devices and you would get the blame. Ultimately you end up making "do I want to break this group of devices or this group of devices" decisions. You definitely aren't starting from scratch, partially because of patents butt also because the reference implementation is the only one which encodes all those legacy nonconforming variance to the spec.
Joel_Mckay 2 hours ago [-]
C is king on mcu with C++ for higher end chips, but most use a subset dialect compatible with janky compilers on multiple platforms.
The age of the legacy mcu is closing, as full blown SoC with specialized peripherals now cost about the same.
It is shite pay, naive people incorrectly assume they can write for real-time, and unless you wrote bare metal Assembly at some point in your past... you will have difficulty understanding why certain things break. esp32 are a cheap capable chip to learn on, and the Pluto course work is a great intro to the Zynq 7010 fpga hybrid Soc.
The world needs Erlang/Elixir OTP devs right now as everything shitty is burning down in CVEs. Easy money if your work can take the "AI" audits and survive. =3
AeroNotix 1 hours ago [-]
Can you expand on why the world needs Erlang/Elixir OTP devs? I have a decade of experience working with Erlang/OTP. I've taken a few months off and the thought of trying to get into SWE work with the AI slop that you've all gotten into fills me with dread.
Joel_Mckay 22 minutes ago [-]
>Can you expand on why the world needs Erlang/Elixir OTP devs?
Primarily the transparent fail-over multi-host cluster coordination areas need cleaned up, or will likely face resiliency issues moving forwards.
>trying to get into SWE work with the AI slop
OTP languages remain operational even when other areas are under active RCE attempts. These systems usually remain operational when pinned at 100% CPU usage by a code defect, interpreter has live internal state debugging as part of the API, and are compatible with constraint programming paradigms (Erlang's list comprehension feature is somewhat functionally similar.)
The fact is daemon functionally will be more important than perfection when repo code is audited several times an hour... long before patches can be pushed out to the public.
"LLM" cannot help much in my problem domains, but are good for looking up poorly maintained documentation. Anything that lies to you 30% of the time is a waste of time, but does certainly bring out the petulant and naive. lol =3
gzell 3 hours ago [-]
I transitioned to web development after 20 years as an embedded developer because that's where the money was at the time. While I did cash it, it was at the expense of good software engineering skills. Maybe worth it but I have my regrets. It's harder to make the transition in the other direction, to be a "real" programmer as we like to call it. Embedded highly encourages good software engineering knowledge and practiceBut all the advice on the comments is good and it's well worth trying, especially if you have an engineering mindset. Good luck.
aman179102 43 minutes ago [-]
[flagged]
benlandseer264 51 minutes ago [-]
[flagged]
VanryxPay 36 minutes ago [-]
[flagged]
victorylaunches 4 days ago [-]
[flagged]
learntoplaypitc 4 hours ago [-]
[flagged]
Xinti 1 hours ago [-]
[dead]
monster11 4 days ago [-]
[flagged]
Theophane_Rame 4 days ago [-]
[flagged]
zergrush 3 hours ago [-]
Far too late at this point because of LLMs which can do embedded/firmware now. I would strongly advise to look for jobs in another field at this point or start designing and distributing your own products.
C++/Rust/C these languages are no longer gate keepable and LLMs can write better code at greater volume and speed than a human can, so the question for a business is they would not need as much engineers, so you are competing against incumbents who are already likely in a position to be entrusted for managing LLMs when they are not looking to expand their teams, makes sense ?
The days of "renting out" your coding services is over, it's similar to how we used to have yarn spinsters that insisted on finding work even as automated systems took over in the UK industrial revolution.
dominotw 2 hours ago [-]
very original comment. This is year 2023 and ive never heard this before at all.
zergrush 17 minutes ago [-]
we've come a long way since 2023 the amount of progress has been tremendous and if you are not keeping up then you are already far too behind to catch up as the gap between those embracing agentic coding vs hose who shun it are widening as are the distribution of opportunity skewing away from those that resist AI
faangguyindia 4 days ago [-]
You can embed quickjs then write js.
Rendered at 19:37:59 GMT+0000 (Coordinated Universal Time) with Vercel.
ChatGPT is an incredible engineering mentor. You can copy/paste chunks of your KiCAD schematic into the chat and it will just jump in and start recommending parts that meet the moment. Just ask it everything, and keep asking follow-up questions.
Make no mistake; you aren't just learning embedded development... you're learning electronics. That means you should get into the mindset that you'll be doing a lot of skill stacking.
If you don't know how to CAD things in Fusion and 3D print them, fix that. I'm happy to recommend an excellent tutorial video.
I recommend getting a bag of 25 of each of the ~100 most common resistor values and putting them in a bin. This will cost ~$100 and you will have all of the resistors. I cannot tell you how great this is. It's like having a jar of every spice.
Amazon has many vendors that sell plastic snap cases of capacitor values and various diodes. I have a stack of different parts. Again, all of the spices.
IMO the most underrated microcontroller is the ESP32-P4. It's a powerhouse, and a company called Wireless Tag makes a module called WT0132P4-A1. It is the bee's knees. RP2354A is also an amazing thing because it has something called PIO that will make you very happy when the time comes.
Not all breadboards are created equal. Many are terrible. Invest in BB1660.
KiCAD has become really, really good. And JLCPCB is both cheap and flexible.
If you get curious about surface mount - and you will - if you remember one thing from all this, it's to invest a few hundred in a decent STEREO microscope on AliExpress. If you give your brain the ability to perceive depth, you'll suddenly realize that you're a soldering ninja. (A Robin's Tools Pixel Pump doesn't hurt, either.)
I absolutely love the degree of control you can get at the microcontroller level. One of my proudest projects had IRQ handlers written in assembly where I counted instruction cycles in order to guarantee worst-case latencies and well-defined impact to realtime activities in the main loop. There were also some lower-priority tasks going on that were cooperatively multitasked.
Some time later I wrote a JavaScript library that mimicked some of those cooperative multitasking concepts to queue background image downloads and other generic "ITasks" in a manner that maximally utilized available connections (which were extremely limited by servers back then) and allowed on-the-fly reordering. The frontend used it to always ensure whatever image the user was most interested in (i.e. tiny previews, and large version of the result presently hovered over or that we anticipate will be hovered next) came up as close to instantaneous as we could get, and the rest of the large images lazy-loaded in an intelligent order. The result was an experience that felt so much faster than how most other websites operated back then (and ironically even now as bloat has really slowed a lot of sites down), and users loved it.
So the concepts you learn at one end of the spectrum can definitely benefit you at the other.
I have much less fun now that so much of the web has moved to React and other cookie cutter frameworks where the level of control comes at a higher cost (in terms of learning curve, understanding all the layers of abstraction, and space for developer creativity).
In other words, just get in there and start playing with a personal project or two, and trust your instincts and all the experience that honed them!
Get comfortable with breadboards, scruffy command-line interfaces, and hex dumps. You'll need to stare at a block of raw memory, and determine the data structure, therein.
I started with embedded. It's a lot of fun, but you'll spend a hell of a lot of time, preparing the stage. You will be responsible for all the low-level stuff that a modern host OS does.
I'd actually suggest starting with pure Machine Code, first. It will make ASM look easy, and help you to really understand the processes in the chips.
Also: IRQ is The Way. It's a whole new way of looking at threading. There are absolutely no safety nets.
That said, it's been 35 years, since I did that low level work. I suspect the tools have come a long way, since.
Now, if you’re just over, focused on technologies and want to use a technology for the sake of using that technology, probably won’t be as easy.
Saying you want to transition to C, C++, or Rust is like a handyman saying they want to to get better with electrical tools vs plumbing tools or circular saws vs reciprocating saws. They focus on the problem than try to pick the right tool for the job.
Focus on what you want to solve, and then apply the tooling. Solve the same Problems you’ve already been solving, with your new tooling and weigh the pros and cons of both. The obvious choices will come out based on the problem at hand. The jump from web apps to firmware may be a big jump. But if you focus on a problem that interests you, and it appears those techs may be a better tooling for your task. Maybe solve it in your traditional languages first, then re apply the new stack. Or contribute to tools you’re already using, written in those stacks.
My point is, whoever has actually done the transition, their experience may not really help you a lot. They might have a cousin who actually refers them to the company; They might have worked 10 years of embedded projects as a hobby; They might be lucky because 5 years ago there were a huge opening somewhere. I heard of a metaphor about elevators -- Several people ride an elevator to floor 100, and a reporter asked them how they get there -- one of them said he did 1,000 push-ups, another said he read 2 books, and a third said he bumped his head against the elevator's walls.
Those experience might be very well relevant, or not. But above all, copying others' experience, in THIS market, probably doesn't work. Now that you are 6 years senior than me regarding SWE experience, I think you already know what to do -- just do it.
I clicked on the title to write a comment saying exactly this. I didn't know anything about the experience or skills. Unless you're trying to get into Enterprise Java without knowing what a bean is, just doing it usually works.
Can't you just learn what a bean is on the spot as well?
If your question is instead how you make a career transition to embedded/firmware work, that's much trickier. The obvious concern is AI's impact on the job market, but I'm chiming in to observe that this isn't the biggest problem with a web->embedded transition. Good jobs doing embedded stuff, at least for people who don't already have a pedigree doing that work, has for as long as I can remember been a bottleneck for people wanting to do this stuff. There's simply a lot more high-level work to be done than low-level work.
I can see bright spots potentially ahead; I'm not a big Jevon's Paradox person but one kind of clearly Jevon-y area is hardware, because we've zorched one of the biggest impediments to getting new projects off the ground. But it hasn't happened yet, and might not.
Couldn’t agree more. I just had to redesign a PCB from an old client and it took me less than a tenth of the time and money that it took me to make the original fifteen years ago. What used to be $30,000 for ten assembled boards is now $2,000 shipped. Finding all the footprints and reference designs was faster than ever, what took two months end to end took a week (not including shipping wait times).
HALs and vendor libraries are in a far better state so writing the firmware felt less like slogging through a MCU reference manual and more like actual programming. LLM to do the brute forcing so that peripherals like the DMA channels play nicely together, which always burned up so much time during bringup.
Hell I was tempted to just through ESPHome on the firmware and use one of their modules with some custom C. I was dreading the redesign because I hadn’t designed any PCBs in a while but I had an absolute blast.
Just start with some cheap ESP32 boards from Amazon, a breadboard, and a wire terminal breakout for the ESP32
QFT
--former(?) embedded engineer who moved into devops
Absolutely true. I'd say it's different by a factor of 10.
But on the flip side, there are a lot fewer people with the skills and/or willingness to do embedded. So it works out to pretty steady employment, at least in my career. Be aware, though, that the smaller market can mean more time spent looking before you can find a job.
By the way, this course was hands down the most hardcore thing I ever did. At one point I had to count high pulses for a number of c for loop ticks to read a temperature, in the process of writing the driver by hand (no AI yet), ending up with too high readings, and I couldn't debug it since it's actual pulses of electrons in the wire, once you signal it to start you can't stop it. In the end it was an off by one error, but not the usual kind, I shifted left 1 bit too much. I got stuck like 2 hours in the library trying to figure out what went wrong (I always suspect it's the hw fault, naturally), so that aha moment was really fun!
There is something special about creating something physically. I wish you make it soon and have fun!
Embedded development has never had a lower bar to entry. There are cheap development boards, free tools, and modestly priced test equipment. Doing embedded development is entirely up to you.
Getting someone to pay for your time doing it is just the generic problem of getting someone to pay for your time doing something you would prefer to do. It depends on other people and your social capital
Who you know, how good a hang you are [1], the economics of the field, your opportunity cost and compensation requirements, etc. are what matter most. Luck aside there’s no free lunch. The place to start is your building your network. That’s hard.
[1] technical skill might or might not be a determining factor within a workplace culture.
But embedded is something else entirely that I'd argue is fundamentally incompatible with the perspective of a web dev. In embedded, you need to be aware at every moment of exactly what the CPU is doing and how many resources are being used. You generally don't have multiple CPUs or even threads. Multitasking means hardware interrupts instead of cooperative threads. You also likely need to be aware of assembly because sometimes you have to write a really hot function in raw assembly to get the required performance.
Embedded is all about resource optimization, and we dev is about throwing more and more resources at the problem. It will probably be a difficult transition to make because it's a fundamentally different way of looking at and thinking about programming.
I would recommend you find a problem, then buy the cheapest microcontroller you can find and spend a few weeks learning how to optimize your code to make maximal use of the hardware while still solving the problem in acceptable time. A big part of getting my current job was talking in my interview about how I solved a novel problem by using an obscure CPU feature in assembly that resulted in 30x fewer instructions than optimized C.
Find a problem, solve the hell out of it, and speak confidently about your solution.
One benefit that web dev does bring you is thinking about concurrency. Your CPU can only do one thing at a time, but microcontrollers have peripherals that run in parallel, and the embedded system usually consists of multiple chips talking to each other and doing independent work. Managing this concurrency efficiently with a single CPU thread is a large chunk of embedded engineering. Instead of IPC, you do things like set up a DMA pipe between SPI and RAM, trigger an interrupt on I2C bytes, use harhardware timers to simulate threads. You'd be shocked at just how much concurrent work can happen with only a single non-pipelined in-order CPU thread.
If you want to break into a new field, the advice is always the same: do the work in your own time, build a portfolio, show you know the job, and speak confidently about the hard problems you learn to solve.
The "hard" stuff you described is already being done with LLMs
For fun - just do it. There are no barriers. Esp32, audio synthesizer gadget, cameras. Everything is cheap and available.
For the money? Expect lower pay, lower standards, worse hours and shit conditions
Transitioning your skills is possible. But embedded is a different world and a different mindset. You a lot less RAM and a smaller CPU. You may have hard deadlines - the right answer can be wrong if it's too late. You may have multiple threads that have to share data and not corrupt it, no matter what order events occur in.
In the worst case, you may not have logging available at all, and you may not have a debugger. (I think the worst I ever had to do was attach a logic analyzer to the address bus, triggered by a specific address, and then use it to look at past memory addresses in order to figure out what path of execution had got me to the point of the crash.)
So, yeah. It's a different world. Learnable, but different.
Getting someone to believe you can do it enough that they're willing to pay you: I don't have great advice there. I spent almost my whole career in embedded. It seems to me that it's easier to get from there into, say, desktop programming, than it is to go the other way. Your best bet might be to emphasize your native development background, and count on there being a shortage of embedded people.
I'm sure you can make Claude help you with this .. but still..logs are important. So is a good debugger and software support for it. It saves so much time in the end.
And in my case, I was actually troubleshooting the code that makes leds blink, with a bit-banged implementation on a deterministic processor.
I recently restructure a logging module in a client’s project, and the resulting performance gains exposed all sort of race conditions.
Sometime all you get is an LED.
This is the reality of highly constrained embedded systems. If you don't like that, I'd suggest moving to web dev where you have state of the art debugging.
You’re also going to have to get comfortable with some level of electronics, because it’s very rare someone wants embedded development done that doesn’t interact with the outside world in some way, and that’s going to require knowing how to drive some IO.
Not having an electronics background or an embedded background, there was certainly a steep learning curve. Lots of stuff to learn and master. But anyone who already knows C and C++, and who is prepared to knuckle down and get stuck in, will do fine.
As someone who frequently hopped around across different domains, the key is demonstrating that you understand their problems and can provide value. Also, being curious helps.
Also, today with AI am not sure how much native development is going to matter. A handful of best practices in an already set up project plus an AI agent is gonna spit out very workable code. Right now the problem space is very much about reasoning about the larger system, de bugging stuff, thinking about how pieces work together, going in the lab and connecting debuggers so that they agent can do stuff to the hardware, etc.
it's already being done without humans, you are already far too behind here
don't know why everyone's acting like this is hard, seems straightforward
I started with data engineering around 2015ish purely because python was easy, and moved into frontend because I needed to make the data understandable for execs. Start building projects and showing them to people, get feedback and iterate. Nothing has changed.
While they are very different, if you know C, it's not hard to transition.
Getting a job in the embedded space though is obviously a lot harder than web. There are simply fewer jobs to be had. Doing some personal projects in the space may reawaken your C skills, and help demonstrate competence.
Good luck!
If part of the reasoning is you are hoping to escape implications of AI I don’t think this will help.
Firmware, especially for the various SoCs, is not a solved problem. There’s far fewer examples of publicly available code for LLMs to learn from, and documentation is at times behind paywalls.
While I’m sure on any long enough time horizon this will change, at the moment finding useful code for some of these SoCs is a lesson in pain. I would also expect relying on an LLM for correct code for low level work like this would be a mistake at best. There are subtle bugs that you can’t shake out with code review and it takes far longer to debug embedded devices than it does web systems.
Have you done any self-directed project to further this ambition? the answer depends on your current state..
Much tech comes along, for example, I think USB-C is brilliant, and I am sure that, if I knew how to make USB-C gadgets, there would be a job there for me, albeit in China, because we don't make anything in the country I am in.
I can use USB-C now, no brain cells required, but nobody is going to give me a job because I know how to plug my phone in. Yet, if I knew the USB-C specs because RTFM, then it would be a whole different ball game, even amongst developers, where they might be adapting reference implementations rather than taking a RTFM + first principles approach.
This list of rules applies to most languages ( https://en.wikipedia.org/wiki/The_Power_of_10:_Rules_for_Dev... )
A bit of history where these rules came from, and why they matter.
"Why Fighter Jets Ban 90% of C++ Features" ( https://www.youtube.com/watch?v=Gv4sDL9Ljww )
The basis of the cited 10 rules helped form many standards for reliable code ( https://www.stroustrup.com/JSF-AV-rules.pdf )
The age of the legacy mcu is closing, as full blown SoC with specialized peripherals now cost about the same.
It is shite pay, naive people incorrectly assume they can write for real-time, and unless you wrote bare metal Assembly at some point in your past... you will have difficulty understanding why certain things break. esp32 are a cheap capable chip to learn on, and the Pluto course work is a great intro to the Zynq 7010 fpga hybrid Soc.
https://www.analog.com/en/resources/evaluation-hardware-and-...
The world needs Erlang/Elixir OTP devs right now as everything shitty is burning down in CVEs. Easy money if your work can take the "AI" audits and survive. =3
Primarily the transparent fail-over multi-host cluster coordination areas need cleaned up, or will likely face resiliency issues moving forwards.
>trying to get into SWE work with the AI slop
OTP languages remain operational even when other areas are under active RCE attempts. These systems usually remain operational when pinned at 100% CPU usage by a code defect, interpreter has live internal state debugging as part of the API, and are compatible with constraint programming paradigms (Erlang's list comprehension feature is somewhat functionally similar.)
The fact is daemon functionally will be more important than perfection when repo code is audited several times an hour... long before patches can be pushed out to the public.
"LLM" cannot help much in my problem domains, but are good for looking up poorly maintained documentation. Anything that lies to you 30% of the time is a waste of time, but does certainly bring out the petulant and naive. lol =3
C++/Rust/C these languages are no longer gate keepable and LLMs can write better code at greater volume and speed than a human can, so the question for a business is they would not need as much engineers, so you are competing against incumbents who are already likely in a position to be entrusted for managing LLMs when they are not looking to expand their teams, makes sense ?
The days of "renting out" your coding services is over, it's similar to how we used to have yarn spinsters that insisted on finding work even as automated systems took over in the UK industrial revolution.