Today's Hardware Design Software Is a Train Wreck — We Deserve Better

Nothing has changed in hardware design software in decades. We’re building AI tools to shake things up.

Sponsored by Zenode
15 days ago

TL;DR - We built a search engine for electronic components powered by AI. We want you to try it here, and send us your feedback so hardware engineers can finally have something nice to play with.

My cofounder and I have spent our entire careers doing rapid turn hardware development, typically going from an idea to a working prototype in ~1 month. In the movies, this would have been like the montage in Ironman where Tony Stark is in the cave building his first suit. But in reality, 80% was just sitting at our computers, reading datasheets and clicking around in CAD software while softly swearing under our breath for 12 hours a day.

That’s because today’s hardware development is littered with soul-sucking, repetitive tasks. Want to extrude that sketch? Too bad, it's underconstrained. Need to shift that trace to a new layer? First you've got to fix all these polygon errors. Need to generate a package for manufacturing? Hope you didn’t miss a note in the fab drawing!

Today's hardware design tools are stuck in the late 90’s, and their makers act like we should be happy to pay them thousands of dollars for something that looks like the UI was designed in Microsoft Paint.

Meanwhile, our software engineering brethren are hellbent on automating their entire ecosystem. Need to stand up a new server? 15 years ago, that took a week and the setup of physical hardware. Today, I can stand up a new machine on AWS in less than an hour. Need to read some random file format? There’s a python library for that and a co-pilot that can write you the interface. No more learning the intricate details and spending a few days writing C++ to extract the data you need. Want internet access in your product? 10 years ago, you needed a full time software engineer to get the network stack running on your MCU. Today you select an option in your RTOS kernel build and it’s done. In all of these cases, the boilerplate has been automated away so you can spend time doing something that matters to the application.

Why are we hardware engineers ALWAYS left with hand-me-down tools while our software brethren get all the shiny new toys? There are dozens of startups offering AI coding autopilots that have raised in the last year, meanwhile I'm over here hitting CTRL-F and copying resistor values out of a reference design’s PDF like it’s the damn Stone Age!

Here’s my theory: Software can afford to screw up. Look at that recent CrowdStrike hack, a catastrophic global breach because some engineer made a mistake. Within 24 hours, code was rolled back, systems rebooted, and everything was fine again.

Can you imagine making that level of a mistake in hardware? Wait, you don't have to; the door blows off a single plane, and within 24 hours Boeing’s stock is in freefall, the CEO is handing in their resignation, and the FAA has grounded the entire fleet until the root cause is determined and a solution devised. Different stakes, am I right?

But for that exact reason, shouldn’t we have BETTER tools than our software cousins? Shouldn’t we have automated systems catching those "oh-no" moments before they snowball into disasters? Why are we still burning hours reading datasheets to hook a circuit up to an off-the-shelf AC/DC power supply? Why is using a red sharpie on a printout of a schematic still the primary method of design review? Where is our continuous-integration equivalent to replace the endless emailing of design_v3_final_thisone_butactually.zip?

Hear me out, AI can and will solve all these problems. Not tomorrow, I’m not delusional—I wouldn’t trust today’s AI to get the color code on a resistor consistently right, let alone route my board, but we have to start somewhere!

And AI isn’t going to make hardware engineers any more obsolete than CNC eliminated machinists; instead it’s going to give us super powers. We won't be spending 90% of our time struggling with the software interface, or mindlessly copying circuits from reference designs, so instead we can focus on the overall architecture and implementation, i.e. “the fun parts”. That in turn will cause a Cambrian explosion of cool hardware being engineered and released into the world!

That's the dream behind why my cofounder and I started building Zenode. We've spent the last year building an AI model that can read the component datasheets, the instruction manuals behind using any electronic component in a design. Spoiler alert, it was really hard; every manufacturer structures things differently, there are tens of millions to consider, and every datasheet is the blend of text, images, footnotes, equations, and charts that LLMs really struggle with.

We'll spend years perfecting this model and all of those built on it, but it's finally good enough to put to work. The first thing we've built is a search engine, because we're done with spending weeks of our lives scrolling through gigantic distributor part catalogs and reading dozens of datasheets before selecting each part. It’s absolutely wild that companies routinely drop millions of dollars on the parts we've picked out, but the only 'aid' we have during selection is CTRL-F in Adobe Reader.

An AI search engine is certainly a far cry from ‘Jarvis’, but it’s just the first step while we improve the accuracy of our more advanced tools. In time, we envision a co-pilot that reviews your design as you draw it to catch little gotchas before you get too far (i.e. Did you know that you’ve flipped the TX and RX pins on this UART?). We’d love to build a tool that takes boilerplate reference designs and customizes their passives to fit our needs (I would die a happy man if I never again had to calculate the compensation passives for a second order filter). Pinouts, power budgeting, footprint generators, there are hundreds of monotonous, tedious tasks that AI can tackle that would save hardware engineers time and frustration, and ultimately help us build hardware better, faster, and cheaper.

It’s definitely still early days here in the AI revolution, but it brings with it the palpable sense that we’ll soon have better hardware design tools. After all, as they say, the journey of 1,000 miles begins with a single footprint! 😉

Brandon Bourn is a co-founder of Zenode, an AI search engine for electronic components.

Latest articles
Related articles
Latest articles
Read more
Related articles