Optimization of high-precision DC voltage measurement

A colleague once asked, “How do I measure microvolts in my test?” High-precision DC voltage measurements can be complex. When measuring, time is money. Therefore, achieving fast and accurate measurements has always been a challenge.

Traditional optimization techniques have used high-precision amplifier circuits and faster measurement devices. These two are still necessary conditions for achieving the best measurement in the shortest time, but they are not sufficient. The inverse relationship between settling delay time and signal noise depends on the equivalent noise bandwidth of the measurement device's driving circuit. The device under test (DUT) and the measurement instrument define the system characteristics, which closely link the settling delay time and broadband noise.

If the circuit bandwidth is zero, the noise will also be zero, and we may be able to measure with one sample, but the circuit will never settle and the DC error will be 100%. Therefore, too low a bandwidth will result in too long a measurement time. On the other hand, if the circuit bandwidth is infinite, the settling delay time will be zero, but the broadband noise will also be infinite, and we will not have enough measurements to average. Therefore, the faster the amplifier, the time required to measure voltage with high accuracy may actually increase.

Let's explore this relationship. In the test sequence, the output of the DUT must settle to within a predefined error range after a voltage step. Assuming a unipolar step response, the settling time will be directly dependent on the bandwidth.

Every voltage measurement includes broadband noise from the DUT, amplifier, and resistors. The amplifier contributes voltage and current noise; the resistor contributes Johnson noise. Since the filter roll-off is not infinitely steep, the noise becomes less important in the region after the –3dB cutoff. The effective noise bandwidth takes this region into account.

For a given broadband noise and effective noise bandwidth, the number of samples required is determined by the measurement tolerance. Basic statistics give the average number of samples required to obtain a 98% confidence level for a given amount of noise. This deviation from the average reflects the repeatability of the single DC voltage measurement. There are many issues in achieving high-resolution measurements, and this article cannot cover all of them. Let's discuss the importance of addressing the overall issues.

Settling delay time. If a component in the circuit has a settling time problem, it will increase the overall measurement time. Limited slew rate is a common cause. Small signal settling time is usually used for calculation. Dielectric absorption is a detrimental issue, so filter capacitors must be selected carefully.

Stability targets. These targets can easily be set too small, such as 0.0001%, resulting in a dynamic increase in measurement time. Since the targets are affected by the step size, larger targets should be used when the step size is a small fraction of the measurement dynamic range. It may be necessary to set the bandwidth for different measurement sequences separately.

Error voltage. The allowable error voltage is often set too small for all measurements. Statistics show that if a Student T table value of 1.6 is used, the measurement deviation will be within the allowable error range 98% of the time.

Voltage reference. This may introduce noise. In the case of a D/A converter, this noise may be code dependent.

Wideband noise. Use a high-quality spectrum analyzer to measure the wideband noise of the circuit directly. With the same number of typical circuit noise sources, it is quite tedious and prone to error to perform accurate calculations on paper.

Measurement accuracy and resolution. Test engineering practice generally requires that the resolution of the measurement device is an order of magnitude larger than the allowable error, but in fact it is always assumed that the accuracy and resolution of the measurement device are much smaller than the allowable error in the actual measurement.

Amplifier. Use low-noise op amps in the signal chain. This is a good way to keep the resistor values ​​low, but not so low that the amplifiers create current drive and thermal issues.

Cost of test requirements require optimization of traditional slow, high-precision measurements. This technique allows us to minimize measurement time, saving money and also a test design.

The semiconductor industry is on the verge of 20-bit DC circuit production. The next issue is the need for test engineers with good professional capabilities. (end)