DFM: The Terrible, Horrible, Supply Chain Woes
I religiously check the new product section of various DIY electronics shopping sites to see what new obscure parts people have discovered. These sites can be an inspiration or a source for quickly getting a proof of concept working. The problem comes when you expect these parts to be more than they were designed to be. This lesson was re-enforced for me in a previous crowdfunding project.
The goal of the product was to monitor and regulate humidity via a single fan as cheap as possible. My first choice was the DHT11 temperature and humidity sensor. It was the cheapest sensor I could find and already had open source libraries. My first prototype was able to be done quickly which facilitated collecting data.
The first mistake I made was obvious after reviewing the information from my data logger. I did not factor in the accuracy I needed when picking my part. The application didn’t need a precise absolute value but it did need repeatability. This device would be measuring a high humidity environment for weeks at a time so drift was important. Though it was the cheapest option, it wasn’t the cheapest acceptable option. I next tried the DHT22 since it was a pin for pin replacement. The second prototype board performed much better and allowed for more accurate test data. I was able to use that information to tune my control loops and get a full alpha device assembled.
Though it was the cheapest option, it wasn’t the cheapest acceptable option.
While the alpha device was going through a testing cycle, I started collecting manufacturing quotes to see how much everything would cost. This is where I discovered my second mistake. I assumed my reasons for using this part aligned with what this part was designed to do.
My application needed a low cost, good enough solution, that could support thousands of production units. Instead, I had designed my device around a part that was optimized for bread boarding, bigger than I needed, and most importantly, unreliable to source in large quantities.
Easily prototyped qualities are great for people who want to quickly wire up a first article, but end up just adding cost to production units.
My third mistake was not checking the supply chain until the end. I didn’t like the price increase of the DHT22 but I was hoping for a good surprise once I went to source large quantities. Since they were available on multiple sites they should be easy to find, right? Not so. After gathering competitive quotes from various US and Chinese vendors, the prices for the sensor kept on getting higher and higher. The available supply was low so the price reflected supply and demand. At the time of launch we had a big question mark around what the part would cost. We decide to kick the can until after the campaign ended since we had a few options. The margin could take a hit margin with the ~$4.00 DHT22 or we could hope that we could magically make the DHT11 work well enough with software.
The Re-Design
Unfortunately the crowdfunding campaign failed, but we still wanted to keep going. This was the beginning of a re-design to further cost reduce and optimize for low quantity batches. Rather than asking for a lot of money to make one big batch, we hope to use the marketing traction to slowly sell much smaller batches over time. The parts of the electronics that can be reduced are the 7-Segment display, input protection, and humidity sensor. Switching the display from white to red is an easy 40 cent win, the risk of removing input protection is a conversation for another day, and the humidity sensor needs improvements.
Toward the end of our development phase a new humidity sensor, SI7006, showed up on the popular parts websites. It was obvious that the price and size were going to be a substantial cost savings. The part is ~$2.25 for a single unit which was already almost half our DHT22 pricing. Since it is a small surface mount part it removed an through hole assembly step and gave me the option to save even more by reducing the PCB surface area. The major down side is that it is a 3.3V I2C part and my initial design was based on 5V. This meant I had to redo my whole schematic to run everything from a 3.3V supply. In this case my time was considered free so it wasn’t a hard decision, but I could have easily estimated the return on investment(ROI). Luckily I had enough pins to switch from one-wire communication to I2C and all my ICs were able to operate fine at 3.3V.
Target Markets
Though the price and size reduction were ultimate goals, I also gain a more intangible benefit. The new part is made for commercial use. Its main market is other large businesses making hundreds of thousands of widgets. This means that I don’t have to get my parts from one relatively unknown Chinese vendor. I now have access to professional grade logistics tools. The chart below shows that stock is available from multiple vendors. This is something I can never know with using an obscure part targeted toward DIYers such as the DHT22. Having a stable supply also keeps the price consistent so I don’t have to worry if my part cost is going to go up 25% in Dec and drop 5% in June.
Large businesses also expect more out of a part than the DIY community. This gives me trust in the accuracy and performance of the part over time. The manufacturer has a large financial incentive to make sure everyone of their parts meets or exceeds the listed specifications. Since these devices are installed in things like HVAC units and clothing dryers, any problem with the device turns into a large dollar warranty incident. If enough of these incidents occur the cost associated will be pushed back onto the sensor manufacturer. The value of reliability becomes a very real dollar figure.
DIY parts have a much different lifespan. When a part fails in a hand built prototype, I just replace it with a new one. Even if I could determine the cause of the failure, it wouldn’t be worth my time to try and fight for a refund. When nearly all of a company’s customers have that mentality, reliability is less of a tangible concern.
It doesn’t mean that a DIY grade part is unreliable, just that there is less of an incentive to be reliable.
The end of this story is rather anti-climactic. We ended scraping the project and moving on to different ideas. My decision to move to the new sensor was later validated by a community member’s independent research. If I shipped hundred of units with DHT sensors, I would have a bunch of angry customers with broken units right about now.
Even though I wasn’t able to be the next dry-age steak unicorn startup, I learned a handful of tricks and put another completed product notch on my belt.
