Comparison of the Josephson Voltage Standards of the NIMT and the BIPM

Transcription

1 Comparison of the Josephson Voltage Standards of the NIMT and the BIPM (part of the ongoing BIPM key comparison BIPM.EM-K10.b) S. Solve, R. Chayramy, M. Stock, S. Pimsut* and N. Rujirat* Bureau International des Poids et Mesures F Sèvres Cedex, France *National Institute of Metrology (Thailand) 3/4-5 Moo 3, Klong 5, Klong Luang, Pathumthani 12120, Thailand

2 Comparison of the Josephson Voltage Standards of the NIMT and the BIPM (part of the ongoing BIPM key comparison BIPM.EM-K10.b) S. Solve, R. Chayramy, M. Stock, S. Pimsut* and N. Rujirat* Bureau International des Poids et Mesures F Sèvres Cedex, France *National Institute of Metrology (Thailand) 3/4-5 Moo 3, Klong 5, Klong Luang, Pathumthani 12120, Thailand Abstract A comparison of the Josephson array voltage standards of the Bureau International des Poids et Mesures (BIPM) and the National Institute of Metrology - (Thailand), NIMT, was carried out in November 2015 at the level of 10 V. For this exercise, options A and B of the BIPM.EM-K10.b comparison protocol were applied. Option B required the BIPM to provide a reference voltage for measurement by the NIMT using its Josephson standard with its own measuring device. Option A required the NIMT to provide a reference voltage with its Josephson voltage standard for measurement by the BIPM using an analogue nanovoltmeter and associated measurement loop. In all cases the BIPM array was kept floating from ground. The final results were in good agreement within the combined relative standard uncertainty of 2.6 parts in for the nominal voltage of 10 V. 1. Introduction Within the framework of CIPM MRA key comparisons, the BIPM performed a direct Josephson voltage standard (JVS) comparison with the NIMT, Thailand, in November The BIPM JVS was shipped to NIMT, Thailand, where an on-site direct comparison was carried out from 3 November to 12 November The comparison followed the technical protocols for the options A and B of the BIPM.EM-K10 comparisons. The comparisons involved the BIPM NIMT/BIPM comparison 2/23

3 measuring the voltage of the NIMT JVS using its measurement loop where an analogue voltmeter was used as a detector for option A and NIMT measuring the voltage of the BIPM transportable JVS using its own measurement chain for option B. For both protocol options, the BIPM array was kept floating from ground and was biased on the same Shapiro voltage step for each polarity, which was necessary to maintain stability during the time frame required for the data acquisition. For convenience, the BIPM array was biased at the same RF frequency with which NIMT operates its quantum voltage standard. This frequency was changed for each new series of measurement with the option B and was in the interval f= GHz to f= GHz. This article describes the technical details of the experiments carried out during the comparison. 2. Comparison equipment 2.1 The BIPM JVS In this comparison the BIPM JVS comprised a cryoprobe with a Hypres 10 V SIS array (S/N: 2538F-3), the microwave equipment and the bias source for the array. The Gunn diode frequency was stabilized using an EIP 578B counter and an ETL/Advantest stabilizer [1]. An optical isolation amplifier was placed between the array and the oscilloscope to enable the array I-V characteristics to be visualized, while the array was kept floating from ground. During the measurements, the array was disconnected from this instrument. The measurements were carried out without monitoring the voltage across the BIPM JVS. The RF biasing frequency is always adjusted to minimize the theoretical voltage difference between the two JVS to zero and in most cases the BIPM array can operate at the frequency of the participating laboratory. In the case of the present comparison, the BIPM array was always operated at the same frequency as the NIMT array. The series resistance of the measurement leads was less than 3 Ω in total and the value of the thermal electromotive forces (EMFs) was found to be of the order of 80 nv to 200 nv. Their influence was eliminated by polarity reversal of the arrays. The leakage resistance between the measurement leads was found an order of magnitude lower than the usual value and equal to Ω. This is to be investigated in the future. NIMT/BIPM comparison 3/23

4 2.2 The NIMT JVS The NIMT JVS consists of a Hypres 10 V SIS Josephson array chip (S/N 2490E7), a VMetrix JVS 1002 bias source, a cryoprobe which houses the Josephson array chip, and a microwave oscillator. The Gunn diode frequency can be adjusted from 74 GHz to 76 GHz and stabilized using an EIP- 578B counter. The 10 MHz reference signal is issued from synthesizers which are linked to the NIMT cesium clock of the Time and Frequency Laboratory. A nanovoltmeter, Keithley 2182A served as the null detector for the option B of the comparison protocol. Its gain and linearity was determined by direct comparison with the NIMT JVS system. All measurements were carried out on the 10 mv range. The series resistance of the measurement leads was measured to 21 Ω in accordance with the manufacturer specifications. The leakage resistance between the measurement leads was measured to Ω for the NIMT JVS. 3. Comparison procedures - Option B The option B comparison took place before the option A comparison. After the BIPM JVS was set up, the array of Josephson junctions was checked for trapped flux. The BIPM array was then successfully biased at the same frequency at which the NIMT operates its quantum voltage standard: f = GHz. The BIPM JVS offers a large RF frequency band over which the quantum voltage is stable. This flexibility simplifies the measurement process as if one of the two arrays jumps during the data acquisition, the effect is independent on the particular array and the software can deal easily with it. Furthermore, as it is possible to adjust the voltage difference between the two arrays to zero within one to three steps, this contributes to limit the impact of a change in the gain value of the nanovoltmeter during the measurement process. During the time of the comparison, we met some very good stability of the voltages provided by the JVS within only one grounding configuration: the reference potential of the NIMT JVS was brought to the BIPM probe and dewar through the shield of the measurement leads linking both JVS in the measurement setup. All the other possible grounding configurations were tested but instabilities on the array voltages were identified either on the NIMT array or the BIPM one. The array voltages NIMT/BIPM comparison 4/23

5 remained very stable once the best grounding configuration selected. More details on the different evaluated grounding configurations are presented in the Appendix A. 3.1 Measurement set-up The measurement loop operated for the option B comparison is based on the NIMT JVS and associated measurement setup (manual polarity reversal switch, nanovoltmeter and software), determining the voltage of the BIPM JVS as if it were a Zener voltage standard. The BIPM JVS was connected to the NIMT reversal switch. In order to cancel out the linear evolution of the thermal EMFs which exists between the array at liquid helium temperature and the top of the probe where the connections are made (room temperature), the polarity reversal must be performed at the level of the JVS (using the bias source). Therefore the reversal switch was always operated on the same position and not actually used as a switch. The results of several short circuit measurements of the NIMT switch allowed to determine the amplitude of the residual thermal EMFs of e= (7±2) nv. The gain of the NIMT Keithley 2182A nanovoltmeter was measured several times (see Appendix A). However no gain correction was applied to the readings as the NIMT software doesn t allow this type of correction. The nanovoltmeter parameters selected by the software were: 10 mv range, 20 readings at NPLC=1 in one polarity of the arrays, and no filter engaged. The timer function of the nanovoltmeter was engaged by default by the software. The measurements were performed with the polarity sequence +, -, -, + where the ensemble (+, -) and (-, +) represent each one a measurement point. During the measurement process, the BIPM and NIMT bias sources were adjusted manually to obtain a voltage difference less than 1 mv after each polarity reversal. This process lasts a few seconds and the total time required to carry out a measurement point was approximately 3 minutes, including the relaxing time period after each polarity reversal. 3.2 Results of the option B Preliminary measurements Some instability was detected on the NIMT voltage and following some investigations on the status of the array, we noticed a low value of the critical current on a junction (I c =90 µa). The array was warmed up to He gas temperature and cooled down in the He liquid again. The critical current NIMT/BIPM comparison 5/23

6 value was measured to I c = 100 µa. The array didn t need to be warmed up again for the whole period of the comparison. The preliminary result, calculated as the mean value of 25 points is: (U NIMT U BIPM ) = nv with a standard deviation of the mean of 1.5 nv (Fig. 1a and 1b). This result confirms that the NIMT SIS-junctions-based primary voltage standard offers a very satisfactory metrological reliability. Fig. 1a: Nanovoltmeter readings leading to the results obtained for the preliminary result of the Option B comparison protocol at the 10 V level. The red squares are the readings acquired in the positive polarity of the arrays and the blue disks, the readings acquired in the negative polarity of the JVSs. The black line represents the evolution of the thermal EMFs between the two standards in the measurement process. The evolution of the thermals, emf, is obtained by applying the formula emf = [(R+) + (R-)] /2 where R+ are the nanovoltmeter readings in the positive polarity and R- are the nanovoltmeter readings in the negative polarity. NIMT/BIPM comparison 6/23

7 Fig. 1b: Individual results obtained for the preliminary result of the Option B comparison protocol at the 10 V level. The solid line represents the mean value of the 25 individual measurements (-2.38 nv). The dashed lines represent the standard deviation of the mean of the complete series Final result with the option B comparison protocol We noticed that the NIMT measurement setup is arranged in such a way that the negative polarity of the voltage standard to be measured (in the present case, the BIPM JVS) is connected to the negative polarity of the NIMT JVS. The nanovoltmeter is inserted in between the two positive polarities of the standards, the high of its input on the NIMT side. This particular arrangement of the setup can possibly lead to an increase of the electrical noise due to a voltage common mode effect. We tried to improve the measurement setup to see if it was possible to reduce the voltage difference between the two primary standards and the associated Type A uncertainty which can be expected at the level of 1 nv with the use of a typical digital nanovoltmeter. To achieve this goal, the following changes were successively made to the setup: NIMT/BIPM comparison 7/23

8 1- The NIMT switch was replaced with a BIPM very low thermal EMFs switch; 2- The 10 MHz reference signal for the frequency counter was selected from a single output of the frequency amplifier distribution rack. Later on, the reference signal was changed to the internal quartz of the NIMT frequency counter; 3- The gain of the 10 mv range of the K2182A was measured in order to correct the raw readings, to compute the results and to compare them to the ones directly calculated by the software (where the gain correction is not applied); 4- The analog filter of the nanovoltmeter was engaged and the Timer function disengaged; 5- The scope which monitors the shape of the voltage steps and the multimeter HP3458A which monitors the voltage across the biasing leads were both removed. None of these modifications brought a significant improvement neither on the dispersion of the measurement points nor on their mean value. After the option A of the protocol was applied (Cf. 4), we came back to the option B of the protocol at the end of the comparison. We decided to power the whole equipment through an isolation transformer and a dedicated metrological earth potential was selected, different from the laboratory one. We checked again that the selected grounding configuration was the only one which brought sufficient stabilities on both arrays. The original nanovoltmeter (K2182A#1) was replaced by another one (K2182A#2) which was found to bring less electrical noise in the measurement loop (Cf. 4). Using this nanovoltmeter, the simple standard deviation of 20 readings was reduced to 8 nv while it was 10 times larger with the previous nanovoltmeter. The nanovoltmeter was inserted between the two negative polarities of the standards in order to avoid the common mode voltage issues. We carried out 25 complete measurements in a single measurement series at f= GHz. The mean value is: (U NIMT U BIPM ) = nv with a standard deviation of the mean of 0.99 nv (Cf. Fig. 2). We tried again different grounding configurations and found out that the grounding configuration operated to obtain the preliminary result was still the only suitable one. NIMT/BIPM comparison 8/23

9 Fig. 2: Individual measurement points (black squares) obtained to calculate the final result of the option B at the level of 10 V. The solid line represents the mean value and the dashed lines represent the experimental standard deviation of the mean of the 25 individual measurement points (k=1) Conclusion The preliminary result showed that the metrological behavior of the NIMT primary voltage standard and associated measurement setup is very satisfactory and fully supports the related NIMT CMC s. This preliminary result could be technically improved with the following actions: - The whole equipment was powered from an isolation transformer and a dedicated earth potential was selected as the reference point. Only one grounding configuration of the measurement setup was confirmed to be suitable for achieving good stability for both quantum standards. NIMT/BIPM comparison 9/23

10 - A different nanovoltmeter, selected for low noise, was inserted between the two negative polarities of the standards in order to avoid the common mode voltage issues. These improvements led to the lowest internal dispersion in the data acquisition process. It should be pointed out that some of the grounding configurations led to such a level of noise that the voltages produced by the arrays were no longer quantized and therefore no measurement could be carried out. The preliminary result could be improved but the final result as well as all intermediate results still show a voltage difference between 1 nv and 2 nv which could be explained by a leakage error on the NIMT measurement leads. This assumption is partially confirmed by the results obtained with the option A comparison scheme (Cf. 4). However, the investigations carried out do not allow to fully confirm this assumption. For instance, it would have been very useful to measure the voltage difference between the two standards for lower nominal voltages. Unfortunately, we didn t have enough time for this experiment. If the observed difference resulted from a leakage error located at the output of the filter of the measurement leads, the recorded offset would have been proportional to the nominal voltage. The result obtained from the option B comparison protocol is: (U NIMT U BIPM ) / U BIPM = with a relative experimental standard deviation of the mean (Type A uncertainty) of u A / U BIPM = Comparison procedures - Option A at the 10 V level The changes in the experimental setup needed to carry out the comparison following option A of the protocol were done in a number of steps, to investigate the behaviour of both JVSs, with respect to the 2 nv offset observed for the option B protocol. 4.1 First series of measurements using a digital nanovoltmeter As we suspected a possible leakage error introduced by the NIMT system, we decided to remove the NIMT bias source from the measurement setup. The new setup requires a switch in front of the nanovoltmeter. The role of the switch is to short circuit the detector input so that the BIPM array will be directly opposed to the NIMT array to push it to a very close voltage and Shapiro step. NIMT/BIPM comparison 10/23

11 The switch is then opened and the detector will measure the exact voltage difference between the two arrays. In this configuration, only the BIPM bias source polarity is reversed in the measurement process. This particular procedure was followed when the voltages provided by the array had been identified as stable, so that the array could remain on the same voltage step for a long period (several tens of seconds) [2]. The BIPM software and a BIPM HP34420A (1 mv range) were controlling the acquisition process. The simple mean value of 20 individual points (Cf. Fig.3), obtained from 2 series of 9 and 11 measurements points respectively, gave: (U NIMT U BIPM ) / U BIPM = with a relative Type A uncertainty u A / U BIPM = Fig. 3: Individual measurement points (black disks) obtained to calculate an intermediate result of the option A at the level of 10 V. The solid line represents the mean value and the dashed lines represent the experimental standard deviation of the mean of the 20 individual measurement points (k=1). As this experiment was the last of the day, we repeated it on the next morning to try to check for the reproducibility of the result. NIMT/BIPM comparison 11/23

12 We obtained a second intermediate result with this measurement setup based on 20 individual points. (U NIMT U BIPM ) / U BIPM = with a relative Type A uncertainty u A / U BIPM = A third intermediate result was obtained from a series of 18 measurement points where the HP34420A nanovoltmeter was replaced by the BIPM K2182A which gave a mean value of: (U NIMT U BIPM ) / U BIPM = with a relative Type A uncertainty, u A / U BIPM = In the first and the third intermediate result, the noise floor of the nanovoltmeters was reached and the mean value of the voltage result was very close to zero. From those results, we could suspect the leakage error observed during the option B comparison to come from the NIMT biasing source. However, this assumption could not be confirmed by the re-introduction of the NIMT biasing source in the measurement loop, because of an increase of the noise, a possible offset due to the leakage could not be identified. We decided to move on to the full option A measurement setup, requiring the use of the BIPM analogue voltmeter. This was mainly motivated by the very good stability of the voltage provided by the arrays. The analogue detector would offer a higher sensitivity for electrical noise in the measurement loop. 4.2 Series of measurements using an analogue nanovoltmeter We could investigate a large number of measurement configurations based on this setup and they are all reported in the Appendix A. We firstly checked that the grounding configuration of the setup was equivalent to the best one identified within the option B. 12 different series of measurements were performed comprising a different number of individual points (2 to 13) within different conditions. In the following the most important investigations performed and corresponding observations are presented. However, it is interesting to look at the complete data set as a single ensemble (Cf. Fig.4) where a clear offset of the voltage difference between the two JVSs, independent on the measurement conditions, can be observed. The mean value of the 49 points gives: (U NIMT U BIPM ) / U BIPM = with a relative Type A uncertainty u A / U BIPM = The experimental details of each series are presented in Appendix A. NIMT/BIPM comparison 12/23

13 After the analogue voltmeter was installed in the measurement setup, a zero volt measurement was performed over 5 points which produced the following result: (U NIMT U BIPM ) = V with a Type A uncertainty u A = V. This result provides confidence that there s no major issue with the arrangement of the measurement loop. - The 3 µv range and 1 µv range of the EM-N11were operated and produced similar results. - If the BIPM dewar was not connected to the reference potential, the level of noise was increased such that the range of the N11 had to be set to 10 µv to be able to carry out a measurement. - The NIMT JVS biasing source was removed letting the BIPM array bias the NIMT array (Cf. 4.1). For the use of the analogue detector, both arrays must be on exactly the same Shapiro voltage step. A single step difference would correspond to a voltage difference of 155 µv and would therefore set the analogue voltmeter to overload. This experiment was run for the first time ever in a BIPM Josephson comparison. The results lead us to the assumption of a possible leakage error at the output of the NIMT precision voltage leads. - The NIMT original biasing source was replaced by a backup device of a different type but no improvement could be recorded. NIMT/BIPM comparison 13/23

14 Fig. 4: Individual measurement points (black squares) taken as a single ensemble of 49 measurement points obtained with different configurations of the option A setup at the level of 10 V. The solid line represents the zero volt axis for convenience. The uncertainty bars are the Type A uncertainty (k=1) corresponding to each series of measurements carried out within different experimental conditions. 4.3 Conclusion of the option A measurements We performed a large number of measurements within the option A comparison protocol starting with digitals nanovoltmeters to investigate on the performance of this measurement setup. We were quickly limited by the noise floor (1 nv) of that type of voltmeters and the suspected offset on the NIMT measurement couldn t be confirmed. After the digital voltmeter was replaced by an analogue voltmeter, we carried out 49 measurements points within different configurations and the plot of those points as a single ensemble tends to confirm a voltage offset of about 2.4 nv due to a leakage error on the NIMT measurement leads. The option A protocol didn t lead to any definitive result. NIMT/BIPM comparison 14/23

15 5. Uncertainties and results 5.1. Type B uncertainty components (option B protocol) The sources of Type B uncertainty (Table 1) are: the frequency accuracy of the BIPM and the NIMT Gunn diodes, the leakage currents, and the detector gain and linearity. Most of the effects of detector noise and frequency stability are already contained in the Type A uncertainty. The effect of residual thermal EMFs (i.e non-linear drift) and electromagnetic interferences are also contained in the Type A uncertainty of the measurements because both array polarities were reversed during the measurements. Uncertainty components related to RF power rectification and sloped Shapiro voltage steps are considered negligible because no such behaviour was observed. Type BIPM Relative uncertainty NIMT Frequency offset (A) B Leakage resistance (B) B Detector (C) B Total (RSS) B Table 1: Estimated Type B relative standard uncertainty components (Option B). (A) As both systems referred to the same 10 MHz frequency reference, only a Type B uncertainty from the frequency measured by the EIP is included. The 10 MHz signal used as the frequency reference for the comparison was produced by a GPS receiver at NIMT. The BIPM JVS has demonstrated on many occasions that the EIP-578B has a good frequency locking performance and that the accuracy of the frequency can reach 0.1 Hz. The relative uncertainty for the offset of the frequency can be calculated from the formula: u f = (0.1/ 3 ) (1/75) 10 9 = For the NIMT JVS, the uncertainty of the EIP-578B counter was estimated as ±2 Hz due to frequency instability. It was taken into account based on the assumption of a normal distribution. The relative uncertainty for the offset of the frequency was: u f = (2/75) 10-9 = NIMT/BIPM comparison 15/23

16 (B) If a rectangular statistical distribution is assumed then the relative uncertainty contribution of the leakage resistance R L can be calculated as: u L = (1/ 3 ) (r / R L ), where r is the series resistance of the measurement leads and R L is the isolation resistance. For NIMT, these were measured to r = 21 and R L = The isolation resistance value includes all the cables from the JVS to the DVM. For BIPM, those parameters are measured to r = 3.65 and R L = (C) For the NIMT JVS, a Keithley 2182A served as the null detector, and its gain (G) is The uncertainty due to gain and linearity error is estimated by: u det = (G-1) U det. A large proportion of the detector uncertainty is already included in the Type A uncertainty of the measurements. This component only expresses the effect of the uncertainty of the detector nonlinearity correction. In order to avoid the influence of the gain and linearity error change corresponding to the change in detector range, the voltage difference between the two arrays in the comparison was kept to a minimum by setting both arrays to the same Shapiro voltage step so that the difference voltage (U det ) was of the order of 150 nv, corresponding to the residual thermal voltages and offset of at the input of the nanovoltmeter. Thus, the relative uncertainty due to gain and linearity of detector was then estimated by calculating the voltage error corresponding to the reading of the voltage difference of 150 nv on the assumption of a rectangular statistical: u det = (1/ 3 ) ( ) ( ) / U BIPM = Result at 10 V (option B) The result obtained following option B of the protocol, is expressed as the relative difference between the values attributed to the 10 V BIPM JVS by the NIMT JVS measurement set-up (U NIMT ) and by the BIPM (U BIPM ): (U NIMT U BIPM ) / U BIPM = and u c / U BIPM = where u c is the total combined standard uncertainty and the relative Type A is u A / U BIPM = This result fully supports the CMCs (Calibration and Measurement Capabilities) of the NIMT. NIMT/BIPM comparison 16/23

17 6. Conclusion The comparison was carried out in the NIMT DC Voltage and Current Laboratory where the environmental conditions together with our investigations to find an adequate grounding configuration of the measurement loop allowed to obtain a good stability of the quantum voltages. The NIMT Josephson Voltage Standard is a commercial versatile system which allows a significant variation of the experimental conditions (frequency, nanovoltmeter, biasing source). The electrical noise environment was found very satisfactory as in particular, the voltages provided by the two arrays were very stable during the time allotted to the comparison. As a consequence, a significant number of series of measurements were performed within the option A protocol of the comparison, which unfortunately did not lead to a definitive result. We could even succeed in biasing the NIMT array with the BIPM array only (biasing source of NIMT JVS inactivated) for the first time during a BIPM Josephson comparison. Both the preliminary result and the final result of the option B are very satisfactory. However, despite a lot of efforts, we couldn t find the origin for a systematic error ranging from 1 nv to 3 nv detected within both options of the comparison protocol. A leakage error at the output of the NIMT JVS measurement leads is suspected but could not be definitively confirmed by this exercise. References [1] Yoshida H., Sakamoto S., et al., Circuit Precautions for Stable Operation of Josephson Junction Array Voltage Standard, IEEE Trans. Instrum. Meas., 1991, 40, [2] S. Solve, R. Chayramy, M. Stock, Yinzhu Zhou, Jinni Lee, and Sze Wey Chua, Comparison of the Josephson Voltage Standards of the NMC, A*STAR and the BIPM (part of the ongoing BIPM key comparisons BIPM.EM-K10.a and BIPM.EM-K10.b), Metrologia, 2011, 48, Tech. Suppl., DISCLAIMER Certain commercial equipment, instruments or materials are identified in this paper in order to adequately specify the environmental and experimental procedures. Such identification does not imply recommendation or endorsement by the BIPM or NIMT, nor does it imply that the materials or equipment identified are necessarily the best available for the purpose. NIMT/BIPM comparison 17/23

18 Appendix A This appendix describes the measurements performed in chronological order. 4 November 2015: The BIPM JVS was installed on the day of arrival. The array Hypres #2538F3 was selected, mounted on the cryoprobe and successfully cooled down to liquid He temperature. The critical current was I c =90 µa. We made sure that the neutral and the phase mains distribution was identical for the equipment of both systems. The mains are the same for all the electrical equipment in the laboratory. The 10 MHz signal reference (NIMT Time and Frequency Laboratory) for the Gunn diode RF sources is provided by a distribution amplifier with several outputs. Two different outputs were selected for the two JVSs. In the course of the comparison, we tried different configurations for the distribution of the frequency reference signal: - Use of the same 10 MHz output from the amplifier with a T BNC splitter to feed each frequency counter; - Use of the NIMT frequency internal quartz reference to be fed into the BIPM external frequency reference signal input. We finally went back to the original configuration which appeared to produce the best stability on the locked frequency of the RF signals. 5 November 2015: We started the day with the search for the stable biasing parameters for both JVSs independently at f= 75.4 GHz. This operation is required before the insertion of the systems in the measurement loop. The NIMT array voltage was found quite instable and we decided to measure its critical current using the BIPM bias source and associated scope. The critical current amplitude was measured to I c = 70 µa and we decided to warm up the chip to the He gaz temperature. The array recovered the expected critical current amplitude (I c = 100 µa) and the voltage stability was recovered once the RF power applied to the chip. A first series of measurement was performed after the BIPM array was connected to the NIMT measurement loop through its reversing switch. NIMT/BIPM comparison 18/23

19 The standard deviation of the results of this first series was significant: 17 nv and we realized that the grounding configuration of the measurement loop was incorrect: the BIPM dewar was not connected to the reference potential. This was corrected and in addition, the power supply of the laptop which is involved in the data acquisition was removed from the mains. A series of 25 individual measurement points was carried out as for the preliminary measurement (f= GHz). The mean value was: (U NIMT U BIPM ) = nv with a standard deviation of the mean of 1.5 nv. We tried to find another suitable grounding configuration of the measurement loop knowing that the NIMT cryoprobe and dewar are always referred to the reference potential: - The BIPM dewar and probe were grounded through a direct link to the earth potential. The shield of the measurement lead which connects both JVS in series-opposition was disconnected from the shunt box in order to avoid any ground loop; - The BIPM probe and dewar were grounded through the shield of the biasing cable (the chassis of the biasing source is always grounded); - In this case again, to avoid any ground loop, the shield of the measurement lead was not connected to the shunt box. The first grounding configuration brought too much instability on the NIMT array while the second configuration brought instabilities on the BIPM array. Therefore, the only possible grounding configuration was that the reference potential of the NIMT JVS was brought to the BIPM probe, dewar and equipment chassis through the shield of the measurement leads linking both JVS in the measurement setup. Two more series of 10 measurement points each were performed within this condition at f= GHz and f= GHz. The analog filter of the K2182A was engaged for the second series but with no impact on the results, which are respectively: (U NIMT U BIPM ) = ( ± 3) nv and (U NIMT U BIPM ) = ( ± 2) nv. 6 November 2015: The NIMT switch was operated on its positive position for all the previous measurement. We decided to carry out a series of 10 measurements to investigate on a possible residual voltage. NIMT/BIPM comparison 19/23

20 The biasing frequency was f= GHz for both arrays and the result was: (U NIMT U BIPM ) = ( ± 1.68) nv. In addition to the previous experiment, a short-circuit measurement (0 V on both JVSs) was performed using the switch: we ended with (U NIMT U BIPM ) = (7 ± 2) nv. This result is in accordance with the ones obtained by NIMT before the comparison. This experiment is performed on a regular basis. This offset was surprisingly not visible in the measurement result at 10 V. The gain of the 10 mv range of the NIMT K2182A was calibrated, as well as the 1 mv range of the HP34420A, which we planned to operate at some point during the comparison. An algorithm was prepared to compute the results from all the raw data (nanovoltmeter readings), including the gain correction. The gain correction is not taken into account in the NIMT software which prepares a summary file from the calculation. The content of this file was compared to the results obtained after computation in order to make sure there was no error in the algorithm. The NIMT switch was removed from the measurement loop and a series of 24 points was carried out at f= GHz which led to (U NIMT U BIPM ) = (-1.89 ± 2.4) nv. The scope which monitors the shape of the voltage steps and the multimeter HP3458A which monitors the voltage across the NIMT biasing leads were both removed from the measurement loop and a new series of 16 points was performed : at f= GHz which leads to (U NIMT U BIPM ) = (0.03 ± 5.2) nv. 7 November 2015: We noticed that the NIMT measurement setup is arranged in such a way that the negative polarity of the voltage standard to be measured (in the present case, the BIPM JVS) is connected to the negative polarity of the NIMT JVS. The nanovoltmeter is inserted between the two positive polarities of the standards, the high of its input on the NIMT side. This particular arrangement of the setup might be sensitive to a possible increase of the electrical noise level due to voltage common mode effects. In order to record a possible impact of the arrangement of the measurement setup, we started by reversing the input polarity of the detector in the loop: the high input of the K2182A was connected to the BIPM side. 13 consecutive individual points were performed at f= GHz which lead to (U NIMT U BIPM ) = (-5.5 ± 2.5) nv. The sign of the difference in the quantum voltages didn t change but the amplitude was increased. It turned out that during the measurement series, the NIMT RF source frequency was fluctuating by more than 1 Hz. NIMT/BIPM comparison 20/23

21 Once this issue fixed by choosing another 10 MHz output on the amplifier, we didn t repeat the measurements but decided to change the measurement setup again, by removing the NIMT biasing source from the measurement loop and to modify the setup according to the option A protocol using a digital nanovoltmeter (BIPM K2182A with the analog filter engaged). The first series of measurement performed with this new configuration was quite noisy, (U NIMT U BIPM ) = (-4.4 ± 12) nv at f= GHz. We realized that the BIPM dewar and probe were no longer grounded. Once this error corrected, we could finally come back to the main experiment to investigate the effect on the results from the removal of the NIMT bias source from the measurement setup. We didn t repeat this measurement as the noise level reached could easily be explained by this fact. Both arrays could be biased easily on the same voltage step at 10 V. We changed the K2182A for a HP34420A on its 1 mv range, still controlled by a BIPM software. 2 different series were performed for a total number of 20 points at f= GHz, (U NIMT U BIPM ) = (-0.4 ± 0.96) nv. As one can reasonably expect an internal detector noise floor of 1 nv for the digital nanovoltmeter, the last result tends to demonstrate that the repeated systematic error of 1 nv to 2 nv observed on the results since the beginning could be interpreted as a possible leakage error introduce by the NIMT biasing source as the systematic error tends to disappear when the NIMT bias source was not part of the measurement setup.. 8 November 2015: We started the day with a series of 20 points performed within the same conditions but f= GHz, (U NIMT U BIPM ) = (-1.6 ± 3.8) nv. We changed the HP34420A for the BIPM K2182A (10 mv range, NPLC=5, no filter). As previously, neither the NIMT biasing source, nor the scope and HP3458A were part of the measurement setup: (U NIMT U BIPM ) = (-0.27 ± 0.95) nv. We noticed that the standard deviation of the nanovoltmeter readings in one polarity of the array with this device was 20 nv, where it had been 5 times larger with the NIMT K2182A (K2182A#1) within almost the same experimental conditions (Cf. 7 th of November). It is very likely that the Timer function engaged on the nanovoltmeter for the experiments of the 7 November was the origin of the observed difference. Another NIMT K2182A (K2182A#2) was available and we used it at the place of the BIPM one. The standard deviation of the readings was about 30 nv and the measurement series of 20 individual points gives: (U NIMT U BIPM ) = (0.14 ± 1.9) nv. NIMT/BIPM comparison 21/23

22 The setup was changed to an option A scenario using a digital nanovoltmeter (BIPM K2182A). The NIMT biasing source was inserted again in the measurement loop: (U NIMT U BIPM ) = (-1.8 ± 2.4) nv. In order to try to extract the possible systematic error from the dispersion of the series, we decided to change the digital voltmeter to the EM-N11 which offers a lower 1/f noise floor. The measurement loop is therefore the full option A protocol configuration. During the first series of measurement we noticed an offset on the meter scale that was not cancelled out while reversing the detector polarity. Five measurements were performed leading to the result (U NIMT U BIPM ) = (-2.9 ± 0.9) nv. After a few investigations, the origin of the non-reversible offset on the meter scale was found in the operation of the isolated output of the N11 for recording the data. This issue was solved when the non-isolated output was selected for recording the nanovoltmeter readings and five measurements were carried out which gave: (U NIMT U BIPM ) = (-2.8 ± 0.2) nv. The systematic error was confirmed with the measurements using the EM - N11 by operating either the isolated output or the non-isolated one. 9 November 2015: In order to test the assumption of a possible leakage on the NIMT bias source, we decided to try to remove the NIMT bias source from the measurement setup and to use the analogue detector to measure the voltage difference. This proposal was supported by the point that both arrays could easily sit on the same quantum voltage step, condition required for the operation of the N11. Two series were performed (one on the 3 µv and one on the 1 µv range of the detector) for a total of 19 points (one outlier): (U NIMT U BIPM ) = (-1.7 ± 0.7) nv. From this result the assumption of the leakage was now focused on the NIMT measurement leads. We carried out a 0 V measurement with 5 consecutive measurements where both arrays were set on their 0 V step (RF power on): (U NIMT U BIPM ) = (-0.2 ± 0.5) nv. We then decided to power all the equipment from a dedicated mains line which has a different earth connection decoupled from the laboratory earth we were using up to now. There was no improvement in the result. The 10 MHz reference signal was changed from the GPS signal to the internal quartz of the NIMT frequency counter but this still didn t improve the result. NIMT/BIPM comparison 22/23

23 The BIPM dewar was disconnected from the reference potential. The effect was to increase the offset: (U NIMT U BIPM ) = (-8 ± 0.6) nv. This result indicates the validity of the leakage error assumption. As the reference potential distribution is changed in the measurement setup, the leakage current path is changed and so is the resulting systematic error on the voltage error between the two JVS. 10 November 2015: The detector operated in the measurement setup was still the N11. The grounding of the NIMT measurement setup was improved in making all the connection points tighter. The NIMT bias source JBS 1002 was replaced with a JBS 500 (backup bias source). None of these changes improved the result: (U NIMT U BIPM ) = (-2.7 ± 0.7) nv. If the reference potential is removed from the BIPM dewar, the systematic error is confirmed to a higher amplitude: (U NIMT U BIPM ) = (-5.9 ± 0.7) nv. As we were getting closer to the end of the period dedicated to the comparison, we decided to move back to the option B comparison using the second NIMT K2182A which was found less noisy than the one tested at the beginning of the comparison. This is where the best comparison result was achieved: (U NIMT U BIPM ) = (-0.97 ± 0.99) nv using the option B comparison protocol. This result is considered as the final result. It is obtained after re-assembling the complete NIMT measurement setup with the laboratory equipment and configuration which exhibited the best metrological performances. NIMT/BIPM comparison 23/23