What Every Hardware Startup Should Know About the Electronic Component Landscape

In the case of small orders, manufacturers may wait to start a production run until enough orders have come in to make it worth it, which can mean long lead times for someone who only wants 100 of something

First, please understand that chip manufacturers don't exist to sell hobbyists or startups a few dozen to a few hundred chips per year. They need to sell anywhere from tens of thousands to hundreds of thousands of components per month. That's their reality. The bulk of their existence is devoted to producing lines of components that are, to varying degrees, pre-sold under contract and allocated to specific products and manufacturers.

Each component category has different dynamics. What's true for common grade ceramic capacitors and resistors might not be true for the high-grade capacitors needed in power distribution systems for high speed boards. The realities for pushbutton switches are different than what happens in the manufacture of integrated circuits, sensors or connectors.

Except for real commodity items that everyone uses (SMD 0402 5% resistors) lots of components are simply not manufactured outside of solid contractual arrangements with OEMs. This is necessary in order to plan volume manufacturing and, therefore, reduce costs.

For example, it isn't rare to have a component manufacturer ask an OEM for a product plan and projected volumes before even agreeing to sell them a part. The processor in the RaspberryPi falls under this category. There are a number of chips in consumer products that are not available for general low volume purchasing. You can't even get a data sheet unless you can show that you are going to consume the chips in volume.

The point is that the component business, be it switches or IC's is all about volume.

How do small volume users get parts then?

I am going to over-simplify this and simply say that we get to buy what's left over after fully allocated production runs. If you want to buy a component that is being driven with demand by, say, Apple, you can --and will-- run into "allocation" events. These are points in time when the chip manufacturer is contractually obligated to deliver x number of chips to Apple. If there's nothing left for non-contractual customers to feed on they simply have to wait until the next production batch.

Distributors go out and sell these chips. They book orders and feed the numbers to the chip manufacturer. Most of these orders require that you agree to "NCNR" terms. "NCNR" means non-cancelable and non-refundable. In other words: You own what you ordered and there is no way out of the deal. They do this because the chip manufacturer will aggregate all of these orders and tack on that number to the batch they are running for their contractual customers.

The accumulation timeline is variable and often depends on market conditions. During the economic downturn (roughly 2008 to 2010) thousands of parts went on allocation. Lead times got out to 20, 30 even 40 weeks. Yes, if you picked the wrong part you had the very real possibility of having to wait almost a year to get in on a manufacturing run. Why? Because nobody was buying anything and they couldn't aggregate enough orders to justify making a run.

The other --and far more complex-- aspect of this is the need to equalize manufacturing volume. In order to support their infrastructure component manufacturers would rather have a steady and reasonably uniform level of orders per unit time (say, per month or quarter) than crazy unpredictable spikes. This means that they might very well force allocation or control events on the supply in order to group orders and schedule them as necessary. It makes no sense to, for example, increase costs significantly during a month when an abnormal spike in volume occurs when you can often simply tell the OEM that their parts will come out of the next run in four to ten weeks.

Distributors will only stock components they know they can turn over quickly. Inventory turns per quarter or year is THE king metric in any business that holds anything in a warehouse. If distributors stocked rarely used components in order to be able to provide you with short lead times they'd lose tons of money and possibly cause enough self-mutilation to go out of business. And so, in the context of an entire ecosystem that has to remain healthy through variable market conditions there is no choice but to have delivery lead times that are "impedance matched" to what's going on with the entire pipeline and not the needs of a lonely small to mid-sized OEM that isn't really floating the boat.

I'm interested in what leads you to believe that.

> other than "Don't pay full retail"

I don't understand what this means in the context of business to business deals. Also, sometimes component suppliers have you over a barrel, and they know it. You'll end up paying way more than whatever retail used to be because you need that component.

Sure. A couple of blog posts and everyone is an expert.

Physical products encompass a very wide range of disciplines. You would be monumentally wrong to think that component purchasing is the most difficult aspect of bringing physical products to market successfully.

It's almost like saying that money is the only variable in building good websites, or the choice of a language or framework. The fact is that software development is a very complex art, often requiring expertise in a wide range of disciplines.

Yes, I know, in my prior post I said "software is easy" and now I am saying the exact opposite. I meant it in relative terms. An individual or a team with the right skill set can hack away and create a world-class software product with free tools, in their bathrobes, in a dorm room. They can deploy it to thousands of people, modify, fine-tune, fix and even pivot the product with little if any cost. College kids right out of school (or even before) can develop and offer significant software products.

Hardware or physical products are not like that. It is rare to have recent grads or drop-outs have the necessary skills to complete such projects. They require a lot of money and multi-disciplinary expertise. These days this generally means expertise in mechanical, electrical and software engineering as well as manufacturing, supply chain management, quality control, testing and more. Depending on what you are doing each of those disciplines often require domain expertise in sub-specialties. For example, electrical engineering ranges from analog and digital design to signal and power integrity --crucial for high-speed designs.

In addition to that, hardware products require expensive specialized tools and often years of expertise using them. Laying out a 12 layer circuit board for an FPGA board with signals ranging into the GHz range isn't something one is going to do with a free PCB layout tool and a handful of online tutorials. The machine I am typing this on has somewhere close to $50,000 in software tools installed --EDA, CAD, CAM, FEA, Field Solvers, Compilers, etc. The cost of the PC itself is a rounding error when compared to the cost of the software, the annual licensing and support agreements as well as the training costs.

Outside of that, depending on what you are doing, you might need tens to hundreds of thousands of dollars in standard and specialized test equipment and tools.

Let's put it this way. YC funds startups to the tune of $15K to $20K to develop the first incarnation of a software product. Just one of our DSO's (Digital Storage Oscilloscope) costs nearly twice that much. The point isn't to boast --we are small potatoes-- but rather to try to impress upon you that the hardware game is very, very different than doing software and the variables involved go far beyond how and where you buy your components or whether your Excel spreadsheet with your BOM is formatted correctly.

I could go on. The point is that in relative terms software products are easy. And that's what I meant.

I once worked for a VP of Engineering who was a hardware guy. I was a hardware guy, things were totally copacetic.

Unfortunately the software guys in that company weren't happy campers, because they were the lowest in the pecking order. The VP was sure that software was "easy", even though he didn't know much about it.

Still, things could be worse. Try working at a company run by marketing (I also did that once upon a time). Then both the hardware and the software guys are sucking hind tit.

It's weird because when we buy off our BOMs from our regular suppliers we'd get well packed components that were easy to goods-in inspect and put to kit. If we needed 250 surface mount resistors we knew we either had a reel, or we would buy a reel because we'd probably use them.

But when the customers supplied stuff it'd come in horrible little bits and bobs. You'd get a strip of exactly 250 SM resistors. (Which is horrible for the P&P machine we were using, and resistors get dropped by the machine a lot). To us it should have been the other way round - "Supply the kit yourself? We charge you 10% (of something) because we still have to goods in inspect it and kit it and etc".

Most manufacturers are happy to do a mix. So, if your product uses one weird IC and you happen to have stock of it, but nothing else, feel free to tell them that you're supplying that part.

Or worst. I personally experienced a sourcing issue kill eight months of engineering design work and hundreds of thousands of dollars. It was a simple case of being a small fish in a big pond and the suppliers not giving a shit about providing advance EOL notices. No recourse. You have to redesign your product and hope to survive.

I have experience of Sage line 50 / line 100. They were awful for the bits most of the company needed to do.

The last part is only semi-kidding. Back when I was a project manager (where I DID have some domain experience), that was my rule of thumb and it was more accurate than I like to admit. In another former life in a different role, we had these 95% schedules which were really 95% confidence assuming nothing significant goes wrong schedules--which were about as accurate as you would imagine.