zyxwvuts
IEEE Instrumentation and Measurement
Technology Conference
Ottawa, Canada, May 19-21, 1997
Ultra low noise current sources
zyxwvu
zyx
C.Ciofi, R.Giannetti, V.Dattilo, B.Neri.
Dipartimento di lngegneria dell' informazione: Elettronica, Informatica, Telecomunicazioni
v. Diotisalvi 2, 1-56126 PlSA (Italy)
Phone +39 50 568538 Fax +39 50 568522 E-mail: ciofi@iet.unipi.it
-
Abstract The current sources which are normally
used in research laboratories are not suitable for
low noise measurement systems because of the
unacceptable level of low frequency noise which
they introduce in the measurement chain. The
most important source of low frequency noise in
such instruments is the solid state device (usually
a Zener diode) which is used as voltage reference.
By using a novel circuital topology in which a
lead battery which does not supply current is
used as voltage reference, we have been able to
design an ultra low noise current source
characterized by a low-frequency noise level some
orders of magnitude lower than that of similar
commercial instrumentation.
The design,
realization and testing of such current source is
presented in this paper,
I . INTRODUCTION.
In low frequency noise measurement systems
( k 1 0 Hz), it is often the case that the device under test
need to be biased at constant current. Traditional solid
state current sources can not be used to this purpose
because of the high level of noise which they introduce.
Sometimes a battery together with a high value series
resistor can be used to simulate a current source.
However, such solution becomes impracticable when
the current which must be supplied is in the range of
hundreds of mA or more.
Moreover, the noise
produced by the battery dramatically increases if the
supplied current overcomes a certain value I,
which
depends on the type and the capacity of the battery [l].
Normally I,, is of the order of magnitude of the current
at which one hundred hours are sufficient to completely
discharge the battery. Finally, it is not possible to
maintain the supplied current constant because of the
discharge of the battery during the measurements. To
overcome the above mentioned problems, purposely
designed ultra low noise current sources must be
developed.
The topology and the principles of operation of a rather
simple ultra low noise solid state current source are
described in this paper. By means of an accurate
selection of the active and passive components, it has
been possible to obtain, in the very low frequency
range, a noise level which is orders of magnitude lower
than that of instrumentation of the same type available
on the market.
II. SYSTEM DESIGN
It is a well known result that batteries behave as low
noise voltage sources provided that the current which
they supply is very low [ l ] . In other words, it is
possible to employ a battery as a very low noise
voltage reference. In fact, in low frequency noise
measurements, a high accuracy in the value of the
supplied current, as that which could be obtained by
means of a Zener diode, is normally not important. It is
much more important to keep the background noise of
the system to a minimum. In the circuit reported in
Fig.1, already proposed in [2], a battery which does not
supply current is employed for biasing a low-noise FET
(8 2SK146 by Toshiba in parallel) used as current
source.
zyxwvuts
0-7803-3312-8/97/$5.0001997 IEEE
-1
\...,.....,'
Fig.1. Schematic of the ultra low noise current source
1486
Authorized licensed use limited to: Rochester Institute of Technology. Downloaded on May 26,2010 at 14:48:29 UTC from IEEE Xplore. Restrictions apply.
zyxwvutsrqp
zyxwvuts
zy
In Fig. 1 RL represents the load. By neglecting the
gate-to-source voltage (a few hundreds of millivolts for
this transistor), the current supplied to the load is given
by :
The value of /GEN can be adjusted by selecting an
approloriate value for Rs. The most important source of
low frequency noise in the previous circuit is the input
equivalent voltage noise source enFET
of the FET.
Therefore, the power spectral density Si,,
of the
current fluctuations in the load is given by :
where SenFET
is the power spectral density of enand g,,,
is the transconductance of the FET. In many cases, (2)
reduces to :
(3)
since, normally, g,R, >> 1.
This circuit, however, has some drawbacks. First of all,
it is not easy to find on the market the active
components which we have used, since they are now
produced by the manufacturer only upon request and,
therefore, only for very large quantities. Moreover, the
simple topology shown in Fig.1 suffers from a serious
limitation : if the power dissipated by the FET increases
(as it happens when a high value of lGEN
has to be
delivered to a low resistance load) the effects of the
temperature fluctuations of the active devices cause a
remarkable degradation of the noise characteristics [l].
However, since the Joule heating depends on the
product IG,N.VD,,
V
,, being the source to drain voltage of
the FET, this undesired effect could be reduced by
maintaining the voltage V,, as low as possible,
independently of the supplied current.
This result has been obtained in the circuit reported in
Fig. 2 in which a different type of active device (two
IF3601 by lnterfet in parallel), easily available on the
market, is employed.
03
Fig. 2. Schematic of the new ultra low noise current source.
Authorized licensed use limited to: Rochester Institute of Technology. Downloaded on May 26,2010 at 14:48:29 UTC from IEEE Xplore. Restrictions apply.
zyxwvutsrqp
zyxwvutsr
zyxw
zy
The peculiarity of this circuit lays in the fact that the
voltage V,, is kept constant by means of the feedback
loop U2-Ul-Q3. The operation of such feedback loop
is as follows : the differential amplifier U2 ( I N A l l l ,
gain=l) reads the differential voltage V,, and provides
the inverting input of the high gain operational
amplifier U1 (OP-27) with a voltage, referred to
ground, equal to VDs. The voltage which is present at
the non-inverting input of U1, obtained by means of
the voltage divider R,IR,, is equal to the desired value
of Vas. The operational amplifier U1 acts as an error
amplifier and drives the transistor Q3 (a Darlington
type TIP122) which sets the value of the voltage at the
drain of the FETs.
At equilibrium; the voltage difference between the
inverting and
non inverting input of
U1 is
approximately zero, which implies that V,, is equal to
the desired value. Such value is about 2 V, and it is
the result of a compromise between the need to
reduce the power dissipated by the FETs and the
need to maintain them in the saturation region.
In order to reduce the effects of thermal transients or
of accidental mechanical shocks, which could cause
slow fluctuations of the output current, an RC low
pass filter (R5, C,+C,) with a corner frequency of
0.8 mHz, has been interposed between the battery
and the gates of the FETs. The values of R, and of
C,+C, (10 MQ and 20 pF, respectively) have been
chosen in such a way as to make the contribution of
the thermal noise of the resistor R,, in the frequency
range above 10 mHz, negligible.
However, for a correct evaluation of SenGEN
we must
take into account the noise introduced by the
preamplifier and the thermal noise of RL.
The power spectrum SK of the voltage fluctuations at
the input of the preamplifier is given by the following
relationship :
svi = Se,,GEN + Se,,,, + R ; s ~ , ,+~ Se,,TH
(5)
where Senpand Sinpare the power spectra of the input
equivalent voltage and current noise sources of the
preamplifier, while the term Se,
accounts for the
thermal noise of R,.
The total contribution S,, resulting from Se,,, Si,,and
Se,
was evaluated by disconnecting the current
source and by leaving RL alone at the input of the
preamplifier. The spectrum SenGEN
was then evaluated
by subtracting S,, from the measured spectrum SK.
The spectra reported in Fig. 3 have been obtained
following the procedure outlined before with two
different values of the supplied current IGEN (23 and
46 mA) and with the same value for R, (200 R). The
spectra have a typical I / f y behavior with a value of y
close to 1.5 in the frequency range from 5 mHz to 1 Hz.
A comparison of the spectra in Fig. 3 with the spectrum
of the equivalent input voltage noise source enFET
of the
FET, which have been independently measured, allow
us to conclude that they are consistent with (3),that is
Si,,, is essentially due to the noise introduced by the
FETs.
zy
zyxw
[A2/Hz]
Ill. MEASUREMENTS
The measurements have been performed by using
metallic film, excess noise free, resistors for RL.
Different values of the output current have been set by
changing the resistors R3, R, (R3=R4)which have the
same function as the resistor R, in the circuit in Fig. 1.
An ultra low noise preamplifier, described in [3] was
employed in order to measure the power spectrum
SenGEN
of the voltage fluctuations across RL for any
provides
supplied current. The measurement of SenGEN
us with a means for the evaluation of the spectrum
Si,,,, of the current fluctuations through the load. In
fact. it results :
~
zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLK
h
~.
\
\
.....................
... .
zyxwv
j
j
i
j
i
~
~
I
0.01
(4)
0.1
1
frequency [Hz]
Fig.3. Power spectra of the current fluctuations at the output of the ultra
low noise current source.
1488
Authorized licensed use limited to: Rochester Institute of Technology. Downloaded on May 26,2010 at 14:48:29 UTC from IEEE Xplore. Restrictions apply.
zyxwvut
The (difference between the two spectra in Fig. 3 are
due to the different values of the resistances (R3,R4)
needed for obtaining the two different values of I,,.
In order to ascertain the dependence of sin,,, on the
DC voltage across R,, a set of measurement were
performed at the same value of the supplied current
(23 niA) but with different values of the load resistance
R,. The value of (SinGEN)”*
at the frequency of 50 mHz
is reported in Fig. 4 versus RL. As it can be seen, for a
giver1 value of the supplied current, Si,,,,, is essentially
independent of the load resistance and, hence, of the
outpllt voltage V,,,.
(Si nrcsl)”’
IV. CONCLUSIONS
The design, realization and testing of an ultra low
noise current source has been described. The most
interesting characteristic of the instrument, at present
used for noise measurements in metal stripes for
integrated circuits, is the extremely low level of the
background noise. The careful design and the
accurate selection of active and passive components,
have allowed us to obtain noise levels similar to those
which were previously possible only by means of low
noise batteries. For a supplied current of about
50 mA, some typical values of the output equivalent
current noise in,,, of the new current source are: 110,
49, 12 pA/(Hz)”’ at 0.05, 0.2 and 1 Hz, respectively.
The excellent noise performances of the instrument
makes it particularly suitable for applications in ultra
low noise measurement systems, where a quasi-ideal
current source is required.
zyxwvutsrq
[pA/Hz ’”1
ACKNOWLEDGMENT
The authors would like to thank P. Frega for the
valuable support during the realization of the
prototypes.
REFERENCES
Fig.4. Output current noise spectral density (f=SOmHz) as a function of
the load resistance. The supplied current is 23 mA.
[ l ] B. Pellegrini,.R. Saletti, B. Neri, P. Terreni, “Minimization of lowfrequency noise sources in electronic measurements” in Noise in
Electrical Measurements, Proc. of 1st lnt Symp. on Measurement
of Electrical Quantities, p. 195-200, IMEKO, Budapest, 1987.
[2] C.Ciofi, M.De Marinis, B.Neri, “Voltage and current sources for Ultra
low noise measurement systems”, Alta Frequenza, vol. 7 , pp. 5558, Luglio -Agosto, 1995.
[3] C.Ciofi, M.De Marinis, B.Neri “Ultra low noise, PC-based
measurement system for the characterization of the metallizations
of integrated circuits”, Proc. of the lMTC96 Conf., A.Barel Ed.,
p.319, Brussels, 1996.
zyxwv
1489
Authorized licensed use limited to: Rochester Institute of Technology. Downloaded on May 26,2010 at 14:48:29 UTC from IEEE Xplore. Restrictions apply.