Discovery and rectification of an error in high resistance traceability at NPL: a case study in how metrology works

Abstract

We broach a seldom-discussed topic in precision metrology; how subtle errors in calibration processes are discovered and remedied. We examine a case study at the National Physical Laboratory (NPL), UK, involving the calibration of DC standard resistors of value 100 MΩ and 1 GΩ. Due to an oversight in the assessment of error sources in the cryogenic current comparator (CCC) ratio bridge used for resistance calibrations, results from the period 2001 to 2015 were in error by approximately 0.7 parts per million (ppm), with quoted uncertainties (k=2) of 0.4 ppm and 1.6 ppm respectively. International inter-comparisons did not detect the error, mainly because the uncertainty due to the transportation drift of the comparison standards was too large to resolve it. Likewise, research into single-electron current standards relied on traceability to 1 GΩ, and did not detect the error because at this resistance value it was on the borderline of statistical significance. The key event was a comparison between PTB (Germany) and NPL (UK) of a new small-current measuring instrument, the ultrastable low-noise current amplifier (ULCA). The NPL measurements took place over one week in early 2015 and involved calibrating the transresistance gain (nominally 109 V/A) of the ULCA. At that time, the transport stability of the ULCA was not well established. Nevertheless, calibrations of the ULCA at NPL using a 100 MΩ resistor were sufficiently discrepant with the PTB calibrations to motivate a thorough investigation into the NPL traceability chain, which uncovered the error. All recipients of erroneous calibration certificates were notified, but their responses indicated that the size of the error did not impact their business significantly. This instructive episode illustrates a positive interplay between calibration and research activities and shows that cutting-edge calibration uncertainties must be supported by a vigorous research programme. It is also important for NMIs to maintain a comfortable buffer (at least a factor of 10) between their claimed uncertainty and the uncertainty that their customers require, so that small errors can be resolved without significant impact on measurement stakeholders.