It’s what’s inside that counts

The goal is not to design a complex project, it is to make sure it works properly. This old piece of wisdom is still valid no matter whether you are designing a sophisticated instrument, a large FPGA, or a huge software product.

Jordan Dimitrov, Electrical Engineer -- EDN, July 29, 2010

The heart of the system was a highpressure, 100W, mercury-vapor shortarc lamp. Its major disadvantage was deteriorating output: At the end of the typical lifetime of 2000 hours, light intensity was less than 60% of its initial value. The instrument had two major approaches for coping with the issue. The first approach used a built-in radiometer to monitor light intensity, and a microcontroller adjusted an iris opening to ensure that output was within 5% of the set value. The second approach was to use the microcontroller to increase the accuracy of the “dose”—the light intensity times the exposure time. An accurate dose was important for a strong bond. Beyond these requirements, customers regularly performed calibration using the system’s built-in radiometer, and our staff normally used a handheld radiometer for preshipment calibration.

The problem arose when a customer bought both a spot-curing system and a handheld radiometer. He plugged the light guide of the curing unit into the calibration port of the built-in radiometer and then into the handheld radiometer and found significant discrepancy in readings. He also complained that readings on the spot-curing display were changing in large steps even at small changes of desired intensity.

Read More Tales from the Cube

It turned out that many things needed improvement. The troubles started with the built-in radiometer. It showed linear response; however, the calibration procedure required only a slope calibration. The radiometer also did not allow for autozero and autoranging adjustment despite the fact that the board had a trimming potentiometer for that purpose.

The second problem was the microcontroller, an Intel 80C196. Although it was a powerful unit, its built-in ADC’s resolution was only 10 bits, and absolute error and nonlinearity were ±3 LSB (least-significant bit) each. The ADC displayed numbers from 0 to 25,000 mW/cm² but provided a minimum display step of only 25 units. The ADC and the controller sections shared a 5.14V power supply, although the firmware assumed a value of 5V.

There was also a 5- to 6-mV drop between the grounds of the internal radiometer and the ADC, contributing one more bit of inaccuracy to the analog-to- digital conversion. In contrast, the handheld radiometer had a 12-bit ADC with maximum error of ½ LSB.

After several hard days and sleepless nights working on a solution, I prescribed two-point calibration for the built-in radiometer, rearranging wiring between the radiometer and the controller boards, using two of the free ADC inputs and corresponding firmware to implement autozero and autoranging for the ADC, and introducing a ground layer for the controller board. I left the low-resolution display as it was. The only way to fix it would have been to use a high-quality external ADC, but I lacked the resources for this approach. Even without this fix, the 5% error threshold dropped from 5.14 to 0.33 W/cm².

I heard later that one engineer had designed both the hardware and the firmware of the instrument. With limited time, he hadn’t had the chance to verify performance and fix problems. Years later, we had to do it in even less time to meet the deadline.

This experience reminded me of two lessons: When designing an instrument, thoroughly do your homework. When buying an instrument, remember that appearance is important but what really counts is what’s inside the box.

---