The a-Si:H samples for this study were prepared by
rf glow-discharge decomposition of silane and phosphine with a gas phase doping
level of 
[PH 
]/[SiH 
].
The temperature of the Corning
7059 glass substrate during deposition was maintained at
516 K at 0.27 W/cm 
power density, yielding a growth rate of approximately 1 Å/s.
Two coplanar conductance channels were
patterned on the 1 
m thick film on the same substrate via photolithography:
sample A was 330 m wide with 810 m between evaporated Cr electrodes,
and sample B was 200 m wide by 950 m long. The contacts were
found to be ohmic up to the highest bias applied, and four probe measurements
conducted using a Keithley 220 current source and Stanford Research 560 voltage
preamplifier demonstrated that the metallic contacts did not induce the
measured fluctuations. Spring loaded BeCu
clips both made electrical contact to the sample substrate
and held it in good thermal contact with the copper heating block.
The measurement
procedure was to anneal the sample at 435 K under oil-free vacuum
to remove surface adsorbates and
defects produced by prior light exposure, [20, 21]
although as will be seen below measurements on unannealed samples
could also show interesting behavior. The initial series of measurements
was performed in a dry N 
atmosphere
after the annealed sample had completed a slow cool over two hours
to 310 K. The chamber was purged
at a rate of 0.6 l/min in order to minimize the partial
pressure of surface-active impurities such as water. The sample was then
maintained under these conditions over the course of a month without light
exposure or reannealing. In later measurements
carried out after the sample had aged in a dry atmosphere for five months,
the sample was cooled to 310 K at a rate of 6 K/min from annealing at 435 K
for 30 minutes before initiating the dry N flow. Lead-acid
batteries provided a steady voltage supply free of 60 Hz pickup with intrinsic
noise beneath the level of the current preamplifier input noise at all ranges,
and two different configurations were used for the measurements depending
upon the magnitude of the fluctuations relative to the d.c. signal, as shown and
described in Fig. 1.
Second spectral analysis required the acquisition of
an individual data set consisting of
at least 
discrete samples of signal fluctuations
in a single continuous data stream so
that meaningful information could be gleaned at the lowest experimentally
accessible beat frequencies.[15] This was accomplished
by feeding the a.c. coupled output of the
preamplifier monitoring the current fluctuations (DL Instruments 564 current
preamplifier)
into the Hewlett-Packard 35670A
dynamic signal analyzer in the time capture mode. The 16-bit A/D
converter in the HP 35670A
discretized the signal and wrote the anti-alias filtered data
to the 8 Mbyte buffer until the desired number of points was recorded.
This arrangement enabled the acquisition
of continuous data sets of up to 1024 consecutive blocks of 4096 points,
each of which would yield a 1600 bin power spectrum when fast Fourier
transformed.
The data was then transferred to a workstation with 1 Gbyte of RAM for analysis.