As an application engineer, I spend a lot of time visiting customers and reviewing their designs. I’m surprised how often the fear of failure guided their selection of analog-to-digital converters (ADCs). The result is a system that meets specifications but does so with significant extra cost.
Being a design engineer means lots of decisions. It’s like being a baseball player except you have to bat at least .975 just to get your contract renewed. Having to make hundreds or thousands of decisions for each design, engineers have come to rely on many “rules of thumb” (ROTs). These can be very handy things.
For example, four-layer boards are better than two-layer boards for reducing noise and electromagnetic interference (EMI). Make power connections four times the width of signal lines. Keep analog lines away from digital lines. Capacitors should be rated at least twice the maximum voltage. And, sample rate has to be twice signal bandwidth, but three times or more is preferable.
An ROT, developed from valid engineering principles, can be very useful. ROTs based on experience can also be useful:
• A new component is not real until there is one sitting on my lab bench, working as specified.
• Some ROTs can’t help but give better performance. However, this performance is at the expense of extra cost.
• Always use NPO capacitors.
• Blue solder mask is the best color to use.
• Components with values containing “47” are lucky and help bless your design.
• Gold-plating circuit boards will keep the bubonic plague away. (I know this to be true because I use gold on all my boards and I have never seen a case of plague at work.)
ROTs developed from fear can even cause you to spend even more than you need and never know it—for instance, no one ever got fired buying as least as much ADC as was used before. Rules like this are developed out of fear of failure and can have a significant impact on cost.
When I got to college, I went to the engineering orientation, where they told us to look left and right because at least one of us wouldn’t be there at graduation. Next, they showed a film of the Tacoma Narrows bridge whip around in the wind and fall apart. Then they stressed that as engineers we needed to take our responsibility seriously because if we made mistakes, people could die! (Great, let’s give the kid some pressure...)
What are the other downsides of taking a risk? Well, the project could fail, you could lose your job, your kids might not be able to get braces or get into the right schools, or the company might fold. The possibilities are endless.
What’s the upside of risk? For me, it has generally meant a parchment certificate that says I did one heck of a great job and a $100 gift card to Red Lobster. If I was extra clever it may be a walnut plaque with a brassy plate with the go-ahead to take my wife to Ruth Chris Steak House as long as I only order the house wine.
PUTTING RISK INTO PRACTICE
With the upside-downside ratio so drastically one-sided, it’s no wonder engineers quickly become so risk-phobic. Here’s an example of how fear of risk can lead to the wrong solution. It starts in the late 1960s with an engineer that designs a 100-count (6.6-bit) single-slope ADC, built with discrete components.
In the late 1970s, the company decides to update the product. A salesman comes by with an 8-bit single-chip ADC that’s cheaper than the present implementation. More resolution, less cost—there’s no risk there.
In the late 1980s, the company again decides to upgrade the design. However, now there is a new engineer because the old one got laid off. The same salesman comes by and says he has a 10-bit ADC that is cheaper than the old one. More resolution and cheaper—there’s no risk here. (Note that the original requirement of 6.6 bits has mysteriously been lost. Let’s chock this up to poor documentation.)
In the late 1990s, the company again decides to upgrade, only this time it hires a consultant. The same salesman comes by with a 12-bit ADC at the same price as the old 10-bit one, but it also includes a built in 0.1% reference. You only need 10 bits, but hey, it’s the same price and we get a better reference. Besides, 10 bits is one part per 1024, so a 0.1% reference makes sense, especially if it’s free.
Now it’s 2010. The company decides to replace its switches with capacitive touch sensors. It calls us out, and we say we could also integrate the ADC. The company says it needs 12 bits, but it would be really nice if we could provide 14 bits.
How did 6.6 bits end up becoming 12 or maybe 14 bits with a 0.1% reference? The engineer was offered a risk-free solution, and he didn’t look at any other options or the real requirement. There’s a reason why the salesmen drive nicer cars than us. Some of them are evil, crafty trolls in human form that can easily exploit your fear of risk to oversell the solution. You don’t make serious advances by avoiding risks. You make serious advances by taking risks you’re good at taking.