This time, it’s different
Although we are used to cyclical downturns in our industry, the persistent semiconductor shortages we have been experiencing are unprecedented. The causes are many: the pandemic, the Russian invasion, too much demand vs. too little capacity, wafer shortages, and so on. The difference with this downturn is that it’s finally effecting real change in the behavior of design engineers.
For years we have been taught that the best design approach is to implement a minimum-parts-count solution using highly dependable ICs. Today, designers are seeing the benefits of using older discrete technologies that have lots of sources. For example, you might immediately think a ramp generator, differential amplifier, buffer, or gate driver should be integrated circuit implementations. But suppose your amazing proprietary gate driver takes up to 99 weeks lead time, which means you won’t ship your product until you can get them.
While it’s understood that your four-core 32-bit ARM processor SoC can’t be implemented discretely, you might be able to use discrete devices—with backup sources—for your less sophisticated circuit functions. That way, if someone decides to stop making, say, MMBT3604s and MMBT3906s, there are plenty of other suppliers left to choose from.
Discrete devices can be used to switch and amplify; you can use them to create oscillators, buffer I/O lines and more. It is also possible to use small-signal discrete devices like 4148 switching diodes, Zener diodes, small signal MOSFETs, or bipolar small signal devices.
Again, it is impractical to make a microcontroller with discretes, but new designs can contain dual layouts. In this case, only one layout on the PCB is populated and the other is not. If a part becomes unavailable, you can simply populate the alternate layout. This approach is, of course, more expensive—but it’s cheaper than not shipping products for two years.
Industry challenges
Component suppliers must figure out a way to make their supply chains more robust. In general, they are doing a poor job, which has forced customers to find alternative options. While suppliers are fixated internally on tracking data and dashboards and so forth, customers are figuring out ways to avoid being “hung out to dry.” They might even resort to designing their own ICs and having them built in foundries themselves.
Another challenge is that many suppliers are not committed to customer orders. When your backlog goes away after waiting out your 22-week lead time, only to see it suddenly turns into 99 weeks, it means that the supplier sold your parts to a more favored customer. (Nice, just wait some more.)
Problem: “Getting semiconductored”
Imagine a proprietary IC that is integral to your design suddenly disappearing. A VP at a semiconductor company, for instance, might decide to kill proprietary products because they are not meeting the right “dashboard metrics” for the investors. Or because of how they read an algorithm and big data. Whatever the reason, your part is gone. This is called “getting semiconductored.”
Figure 1 is an actual example of this problem. It shows a relay driver coil that was used in a cooling system driving a 75mA contactor. This single part offered many types of protection with one insertion in production and no external parts were needed. Then, with little notice, this nice, compelling part went EOL. (Whoops! Redesign time!)
Solution: Using discretes
What do you do now? Maybe change the design using what you should have used in the first place: discretes. How about a 2N7002? Lots of people make those. And we can add some 1N4148s or 1N4004s and TVS devices, as needed, to clamp the inductive EMF that could damage the device.
The solution is to rebuild the circuit in Figure 1 with “common household items”; that is, the tried-and-true discrete parts we know so well.
Figure 2 shows the discrete design. Q1 is a 2N7002. D1 and the back EMF diode of the relay coil can be a 1N4148 or 1N4004, while D2 can be a BZT52B9V1-G, a P4SMA9.1, or similar. R1 can be approximately 100KΩ, however, it’s not critical. R2 can be 100Ω depending on the logic switching signal applied to its “input” (it’s just there to protect the gate of the Q1.)
And now you have a future-proof design that is built to last. None of the components are critical and they don’t need to be precision devices. Simply build the circuit, times two, and you have a replacement for the original sole-source part. Are there any downsides to this design? Probably not with space or cost. Maybe with board insertions (if you keep score on that), but this is a small price to pay for not getting shut down.
Using this solution, no singular component supplier can shut down your production and force a redesign. If you must use a substitute, it’s easy to find something pin compatible that provides the right function. Plenty of people make discrete devices, and all of us are really good at making them in massive volumes. So, if a product line manager decides to exit the 2N7002 and 1N4148 business (“dashboard metrics baby—and let’s change our logo while we’re at it!”), you’ll still have 50 others to choose from. (Not all are ones you should buy from, but that’s another article.)
Lessons learned
Implementing functions discretely, when possible, means that the circuit, and the overall product, can be produced. You will withstand downturns due to random decisions by executives at the supplier semiconductor company, adverse market conditions, natural and manufactured disasters, and whatever else comes along.
Also, make sure your supplier can answer questions about MTBF and FIT reports, reliability qualification reports, and can provide quality data in production (i.e., AQL levels). If they can’t, look elsewhere.
These types of parts have been around a while and they aren’t going anywhere, so this approach future-proofs your designs. Use discretes and avoid being semiconductored again. After all, fool me once…