Shoot The Engineer

By Luke Lang
Many years ago as a junior engineer right out of college, my manager explain to me the concept of “shoot the engineer.” Engineers are trained to be perfectionists. We want to design the best mouse trap ever. However, the engineer that designs the first working mouse trap takes home the money. Given another day, another week, or another month, we can always improve upon our current design. But will the return justify the investment? At some time, we have to shoot (stop) the engineer from further tinkering and ship the product.

When it comes to low-power designs, I have seen a wide range of simple to unnecessarily complex power architectures. As you can probably guess, the designs with simple power architecture usually tapes out smoothly and get to market first. The designs with unnecessarily complex power architecture usually have lots of problems taping out. Everything and everyone gets blamed, except for the real reason for the problem—the engineer was not shot early enough.

In one of the previous blogs, I wrote that methodology equals restrictions. Today’s low-power methodology is built upon a standard-cell ASIC foundation. Therefore, the low-power methodology inherits all of the restrictions of the standard-cell ASIC flow. The first restriction to keep in mind is that this is not a full-custom design flow. Make sure that the low-power methodology, flow, and restrictions are well understood before trying something new.

Sometimes, there may be good reasons for doing something differently. Make sure that the investment and return are both well understood. Once I worked on a design with extremely complex power architecture. I asked the designer if the return (power saving) was worth the investment (time and effort). He didn’t have an answer. He never bothered to compare his chosen power architecture against a simpler architecture.

Based on my experience, the 80/20 rule works very well for low-power designs—20% of the most common power architectures will work well for 80% of the designs. Start with a basic architecture and do some analysis to establish a baseline. Any improvements on the basic architecture must be analyzed and compared against the baseline. This should quickly lead to an optimized power architecture. And no one needs to be shot.

In closing, I would like to leave you with another thought that I saw on the T-shirts of a youth basketball team. “Good enough never is.”

–Luke Lang is a senior product engineering manager at Cadence Design Systems.