We discussed this subject on the recent Metrology Meeting 2024 in Stuttgart:
The StD or the Allan Deviation is used in DCV metrology (as communicated by Mr. Luis Palafox from PTB) as the decisive parameter of stability.
At first, you need a very quiet DCV source, e.g. a 10V JJ array, which would give about 2nv rms @ 10V, or 2*10-10 noise, about 10 "digits".
The HP3458A already displays 9 digits, if you read via GPIB with OFORMAT DREAL, or if you use Statistics Mean and Standard Deviation.
The performance with usual LTZ, LTFLU based references then is about 100 .. 200nV rms noise, i.e. 0.02ppm or a bit less than 8 Digits.
This is a convolution (quadratic sum) of the source noise and the noise of the DMM.
I regularly achieve such noise figures, and 9 Digits, by taking 16 measurements of NPLC 100 with statistics on the DREAL data with Welfords Algorithm.
Measuring an ADR1000 based reference often gives noise figures below 100nV, i.e. 0.01ppm or full 8 digits.
Mostly the flicker or popcorn noise of the internal LTZ1000A reference kicks in.
Therefore, a more silent reference like the ADR1000 inside the 3458A, and several less noisy OpAmps on the ADC board would improve its noise performance, but you would again need a much less noisy DUT for judgement.
ChuckB, aka Chuck Beuning demonstrated on the MM2024 both:
He presented his new analog ultra low noise DCV reference Z10, having an StD / Allan Deviation near the level of a JJ array, i.e. 4..5nV @ 10V.
Chuck, in cooperation with xdevs, they developed a 4 fold ADR1000 reference for the 3458A, and replaced 3 OpAmps for the +/-12V and 5V reference of the ADC.
Chuck presented the Allan Deviation of the altered 3458A, measuring his Z10 reference, or a zero input on the 10V range, like in xdevs former DMM noise comparison campaign.
For the Z10, he achieved about 40nV rms, i.e. 0.004ppm, and for the short 15nv rms, i.e. 0.0015ppm.
Therefore, the 3458A is capable of delivering near 9 digits stable results with these changes.
Frank