My core design philosophy with electronics is to balance calendar speed, development cost, and risk management. While this might vary slightly depending on the needs of the project, most of the time it starts with a simulation in LTSpice followed by a custom board design.
The bottleneck in most calendar speed issues is being able to iterate. If you analyze how quickly you can iterate, you can judge how long something will take. For instance, in an analog simulation like a power supply or signal conditioning circuit, you can change circuit parameters VERY rapidly compared to modifying component values in a custom board design on the bench. Besides being able to change component parameters, it’s also much easier to measure in-circuit parameters like current and voltage while in simulation. Instead of using a current probe and a wire loop in hardware, you can probe current through a circuit element with the click of the mouse. Therefore, it makes sense to keep the design in this stage until risks are mitigated.
Designing to “Femtofarads”
How long should you keep a design in this stage? It helps to recognize that you’re only going to catch so many errors with a simulation. It helps you get voltage levels and gains in the right ballpark; it helps you estimate if one mosfet is going to have less power dissipation than another (most of the time), but it doesn’t make sense to design to “femtofarads”. I think I first heard of this term from the guys on the Amphour podcast. It’s basically poking a bit of fun at people who design far beyond the point of getting a return on the time invested. For instance, how much time should you take reducing the cost of a circuit that is part of a product with an annual sell quantity of 100? If you can make a $100 circuit cost $10 with another 40 hours of effort, you’ll save $9000 in the first year of selling that product. However, if you’re only going to pull $10 out of that same circuit, you’re only going to save $1k in that first year. Is that worth another 40 hours of design time? If you assume that your time is worth $50 to $200 / hr (depending on experience), that 40 hours of design time could cost $8k, making the extra effort to go from $100 to $90 in BOM cost a bit silly.
Custom Board Design
I typically skip the breadboard phase where you have a bunch of wires sticking out of a protoboard, and DIP components with pins on 0.1″ centers populating rows of holes to stick wires into. This was a lot of fun in college, but trying to troubleshoot one of these designs can be a nightmare. Additionally, most packages these days don’t even come in through-hole packages. How do you breadboard a QFN package? Not very easily.
The turn time for custom boards is much slower than a simulation, but depending on the complexity, the design of a custom board could look like this:
- 1 week – schematic
- 1 week – pcb layout
- 1 week – pcb fab by mfg
- 1 week – pcb hand population or machine placement
- 1 week – bringup
So this quickest you could iterate in a design loop like this would be about 5 weeks. This would apply to something like a switching power supply circuit, but not a complex multi-channel fpga design. These types of projects are typically made up of many smaller circuits, many of which are getting prototyped at the same time. I’ve found that when I control all of these elements myself, it yields the quickest development times. It can get hampered if it takes a week to get a purchase order authorized, a week (or months) to get signoff from different departments in a company, etc. If you really want to sprint to get electronics design done quickly, all of these bottlenecks must be removed from the equation. This is one of the reasons why I like consulting – its just requirements and budget inputs, then sprinting to get things done.
One project I worked on involved a power supply that took input from vehicle 12-24V, and required 4 battery charging outputs and 5 different DC outputs that needed to be able to range from 5-24V at 10A on each output. In simulation, things worked so smoothly. But in actual hardware, it was a different story. That entire board had to be revised about 4 times over the course of a year. But at the end of the day it ran cool as a cucumber. How long would that have taken if it had to work within the framework of a company that was tasked with what I call “protecting the recipe”? Probably 3-5 years. This is where hiring a consultant really shines – it’s the ability to step outside of organizational processes that are designed to protect a mature design, and instead engage a process that is meant to drive innovation, creativity, and results as fast as possible.
Mitigate Risk Upfront
When designs get really complicated, sometimes it is difficult to decide what to work on first. If we go back up to a 20,000ft view of design and look at industrial and mechanical design aspects of a new product, we can look at things like product appearance and usability. These are very important aspects of a design, and might represent certain levels of risk in their own right. But, I’ve seen many projects run out of money when the mechanical design team was tasked with creating elegant, sexy skins with plastic injection molding when sheet metal would have worked fine for a prototype. The money that is spent at each phase of prototype development should be attacking the biggest risks that remain in a design. So, instead of spending $200k in tooling for all the skins of a biomedical product, perhaps it makes more sense to spend $10k for prototype sheet metal parts and work on other parts of the design first which have more technical design risk.
In custom electronics design, this might look like prototyping independent circuits first. For example, one project that I worked on needed over 160 independent temperature controlled zones that needed to be heated and cooled. For this design, it made complete sense to use Thermo Electric Coolers (TECs) or Peltier devices. To get any meaningful temperature deltas in applications like this, it usually involves high levels of current. In this application, we needed to supply up to 5A of continuous current and a maximum of 12V across any given TEC to achieve the heating or cooling power required. 160 separate drive circuits were needed, so I used an FPGA to control the bridge logic for all the drivers. I separated the board designs to use duplicate boards that each drove 20 channels of bidirectional current per board at these levels. In order to keep the cost manageable, I used discrete h-bridges and mosfets. This circuit, along with the control topology internal to the FPGA, was the focus of the first prototype. Mitigating this risk up-front allowed us to show that the product had a chance of seeing the light of day based on budget constraints.
Custom electronics design is not easy. Thus the saying, hardware is hard. Identifying core specs, removing bottlenecks, and allowing top talent to “sprint” allows electronics development to get done in the fastest way possible.