222

IEEE

JOURN.4L

OF

SOLID-STATE

CIRCUITS,

VOL.

SC-8, No. 3,

JUNE

1973

f%

Precision Reference Voltage Source

KAREL E. KUIJK

-k-[
x

__l
v

__._n1

A&s fracfWith increasing temperature the base-emitter voltage of a transistor with a constant current decreases, while the difference in base-emitter voltages of two identical (integrated) transistors having a constant current ratio increases. From the sum of the two voltages a nearly temperatureindependent output voltage is obtained if this sum equals the gap voltage of silicon. A reference voltage source of 10 V based on this principle is described. The reference part of the circuit is an integrated circuit, and thin-film resistors with a small relative temperature coefficient are used. An operational amplifier and a few resistors and capacitors complete the circuit. The source has a parabolic temperature characteristic and the temperature peak can be controlled by resistor adjustment. A change of + 10 K in respect to the peak temperature causes an output voltage change of 250 LV, while a change of + 30 K causes a chsrrge of 2.2 mV. A long-term stability of 10 ppm/ month was measured. The circuit can compete with the best available Zener diode sources, and has the added advantage practically no selection is necessary. that

i
&
Fig. 1. Fundamental circuit.

(2) In the case of integrated transistors the ratio of the saturation currents is equal to the ratio of the emitter areas and hence temperature independent. The second term in (2) will consequently increase in proportion to temperature if the resistor ratios RZ/Rs and RJRI are temperature independent [4]. The first term in (2) shows an almost proportional decrease with temperature:

1. I~TRODUCrIO~ N IDEAL dc reference voltage source should supply an output voltage with a very low temA perature coefficient, a great long and short-term stability, a low internal resistance, and should be insensitive to loading. At the present time there are two types of voltage sources, namely the standard cell (saturated and unsaturated) and the Zener diode [1]. Saturated standard cells have a voltage of 1.018 V, stable to within 1 pV/yr. However they have a high temperature coefficient, amounting to 4o ~V/K at 20C. Unsaturated standard cells have a low temperature coefficient of about 5 pV/K, but they have a drift of 20-40 pV/yr. Both cells show temperature hysteresis and cannot be loaded at all. Zener diodes with a low temperature coefficient are made by placing a Zener diode with a positive temperature coefficient (Zener voltage > 5 V) in series with a diode (negative temperature coefficient). By selection, a temperature coefficient of <5 ppm/K can be obtained. Low long-term drift of <25 ppm/yr is possible by selection that requires measurements over a period of some months [1], [2]. The principle of the reference source described here will be explained with the help of Fig. 1 [3]. In this figure D1 and Da are integrated transistors, connected as diodes. It applies
1, = Io,(exp
(qV,el/7cT) (q V,@,/kZ)

-.

where VGO is the gap voltage of silicon at O K. Hence the output voltage is temperature independent if the temperature dependence cancel each other, in which case the output voltage is [5]

v, = Vao
A more exact calculation of the temperature will now be given. From (1) it follows that

(4) behavior

(5) From [6] we have I. = ATn,2P~

ni=

= BT3 exp ( qVGo/kT) (6)

p. = cTn

1)
1). (1)

. :

12 = 10,(exp

where p. is the mobility of electrons, n~ the intrinsic carrier concentration, n a parameter depending on the doping level, VGO the linearly extrapolated gap voltage of silicon at O K, VeO = 1.205 V [6, p. 24], and A, B, and C are temperature-independent parameters. From (6) it can be deduced that Io = DT exp (qVGo//cT), (7)

---I+* L-J-,.

On the assumption that the input currents 1, and 1~ of the operational amplifier are negligibly small, compared to Iz and 11, the output voltage V. of the operational amplifier will be
Manuscript received November 2, 1972; revised Research January 30,

where D = ABC andq = 4 n. Substitution of (5) and (7) in (2) on the assumption that the output voltage is temperature independent at T = T,o,after a few calculations gives Vo=VGo+ (T1)~(7l)f lnf (8)

. ..-.
1072

The author is with the Philips hoven, The Netherlands.

Laboratories,

Eind-

KUIJK

: PREC1S1ON

REFERENCE

VOLT.4GE

SOURCE

223

Writing T = !!o + AT, one finds as a good approxinlation, if AT/TO <<1, that

This means that the output voltage is theoretically a parabolic function of temperature. The output voltage at T = 20is equal to

(v,),.

v.. +

(7 1)

~+

(lo)

b----[

Using the curves in [6, p. 39, 41] it can be deduced that n = 1.8 at the normal base doping level of 5.106 atoms/ cm, hence ~ = 2.2. If this value is substituted in (10), together with VGo = 1.205 V, we find (lo) 3001{= 1.236 V. A temperature shift of AT = 30 K should give an output voltage change of 156 pV. In the next section it will be shown that the value of (V,) ,00,{ is in agreement with the measurements, but that the curvature of the measured parabola differs from the one found with the theory given previously. II. CIRCUIT CONSIDERATIONS AND TENIPERATURE
MEASUREMENTS

Q-.--l.
Fig.

2.

Integrated part of the circuit, consisting of pairs, a preamplifier stage, and emitter followers.

8 diode

a ~ 16 ? .; f2 ; E8 2 4 t

u!

&9.880V*0.006V

Owing to a few sources of error, the circuit of Fig. 1 is not directly useful in practice, One of the major sources of error is the voltage drift of the operational amplifier, at least if this is not a special one chosen for low drift. This voltage drift will cause a shift of the parabolic temperature curve. The influence of the voltage drift can be reduced by the insertion of more diode pairs and a low drift preamplifier, as was done in the practical solution (Fig. 2), where 8 diode pairs and a preamplifier stage plus emitter followers were integrated. This circuit is expected to have a parabolic temperature dependence, with a peak at TO = 300 K if the output voltage is 8 x 1.236 = 9.888 V. If the temperature should vary Y30 K, an output voltage change of 8 x 156 = 1248 pV is expected, according to the calculations in the previous section. From the measurements it was found that the temperature dependence is parabolic with a peak at TO ~ 300 K if the output voltage is (Vo) TO= 9.880 V. The spread of the peak temperature, measu~ed on 31 samples, is given in Fig. 3. A temperature change of A30 K, however, gives an output voltage change of about 2.2 mV. From these values VGO and q can be calculated using (9), which gives n = 3.125 and VOO = 1.180 V. A measured curve and a curve calculated using these values are shown in Fig. 4. These values for ~ and VGO are not in agreement with those reported in the literature, An ex~lanation for the differen~es has not yet been given. The previously mentioned figures for VGO and ~ can be used in further calculations. For instance it can be calculated that a shift in the peak temperature of 10 K will correspond to an extra output voltage drift of 49 pV/K or a shift in peak voltage of 14.7 mV. The measurement given in Fig. 5 shows that a variation of A 15 mV in peak voltage caused a change in peak temperature of about +12.5 and -9.5 K, which is in reasonable agreement with this theory. It also can be calculated that resistors should be used with a. small difference in temperature coefficients (53 ppm/K) while the absolute temperature coefficient

r
r ---F

0102030405060Z2

7
(C)

n
measured on 31 samples.

Peoktemperofure

Fig. 3.

Spread of the peak temperature,

. o

b--=

-7

IIJ

-2.8

10

20

30

m 40 ----- T (OC)

83
using the

Fig. 4.

Measured and calculated temperature curve empirically found values of VGO and n.

o
K

A%
(m I(J :

, ,

-2

X*
-3 w
5. Temperature

,rl

10

40

.50 T (Oc)
at three

60
different

Fig.

behavior of & sample output voltages.

224 thin film netiork r----%---m

IEEE JOURNAL

OF sOLID-ST.4TE CIRCUITS, JUNE

1973

-7s[:
.

_.._n.1
x

L____ --_________ +:FFz___+_______i-_______i =&

Fig.

6.

Complete

circuit

of

the reference

voltage

source

is far less important. A thin-film resistor network is very suitable for this purpose. Furthermore it turns out that the input voltage drift of the preamplifier, which is

-IEF

<2 with The

~V/K, extra

is still drift

an important occurring

source the

of error, rows in of

together diodes. 3 are

-m-::

between

variations

of

the

peak

temperature

Fig.

probably due to this extra drift. Fig. 6 shows the complete circuit of the reference voltage source, designed so as to make all other drift contributions much smaller than the input voltage drift of thepreanlplifier .The supply voltage for the preamplifier is derived from the output voltage. The output voltage is adjusted to its nominal value by means of the potentiometer, in which case the nominal temperature behavior occurs. III. SETTLIATG TIME, hTOIS~,AND STABILITY

min measurement in a bandwidth from dc to 0,07 Hz. A noise voltage of 17 ~VrP was measured that means 9 pVzj foreach circuit. In a bandwidth fromdcto 100kHzthc noise voltage was 100 ~VPP) and in the full bandwidth from dc to 500 kHz it was 200 yVpg. To measure the long-term drift, seven randomly chosen circuits, kept at a temperature of 36.6 C + 0.8 C, were compared for a month with a differential voltmeter (.J. Fluke, model 881 A). A drift of 10 ppm/month was measured for the worst circuit and of O ppm for the best circuit (see Fig, 9), while maximum daily variations of *2 ppm around the average value occurred. IV. OTHER
OCTPUT VOLTAGES

If another output voltage is wanted, for instance 10,000 V, instead of 9.880 V, the method shown in Fig. 10 can be used. It can easily be calculated that

With a view to stability and speed of response to fast load variations, the preamplifier is bypassed with a capacitor Cl, and a second capacitor Cz equal to Cl is connected to ground. The response of the circuit to fast load variations is now determined by the time constant of the operational amplifier. When a pA741 operational amplifieris used with a time constant, = 0.32 ,m (fo~~ = 0.5 MHz), an output voltage variation will have a value equal to 0.1 percent of its initial value after a time 77 = 2.2 ps. Themeasurements given in Fig, 7 indicate that asudden increase in output current of 5 and 10 mA corresponds to output voltage changes of 400 and 600 mV, respectively, which are eliminated in 1,5 and 2.2 ps. This is in good agreement with the theory. The LF noise was measured as the difference between two reference sources. Fig. 8 gives the result of a 20-

v, =

)?2

v,,,

1+ J

R,
R5+R,

()

l+~j

under the condition R R, 2=% If (13) (12)

KUIJK

: PRECIS1ON

REFERENCE

VOLTAGE

SOURCE

225
*VB : floating supply

r-n% 1 M.fk.

I I

chopper

stab. ampl.

A~
(mV)

ref source 1

PHILIPS

., 1
&!---/*

to
-1oo -200 -300

L-_\---oil bath

I=5.4mA
R= 1.8k.O. I

-J=%--l!

400
50 -m o 40 1 2

3
--+ Time(js)

45

(N) to
-1oo -200 -300 .420
500 -m 0 1 2 3 4.5 Titne (p+

/ I= 9.7mA
1?= 1.8/2.2k Q

--+

Fig. 7.

Circuit

for the measurement of the settling the results of the measurement.

time

and

Fig, 8,

Circuit

for the measurement. of the LF noise and the result of the measurement,
v

52 + f

A C/ $.5 :, ( I 1 1 1
( 1 1 0 o~

b--=

%
%

12

I I

I I I

16 ~

20
Time (days)

24

28

-lo
o

I I I I 4 I
I I

I
I I I

I
I I I

I
I I I

I
1 I

I \
I I I I

1 28
chosen circuits.

f2

!6 _

20
T{me (days)

24

Fig. 9.

Best and worst results of a long-term

drift measurement

performed on seven randomly

(11 ) can be written as 1 &klln&& V,el + q I,, R, l+czR, [

V.

SPECIFICATIONS

Vo = rn(l + .)

. 1

Output

voltage (40C)

(14) Temperature dependence Long-term stability Noise: 0.0008-0.07 Hz dc-100 kHz dc-500 kHz Load regulation Settling time

If (14) is compared with (2), it is obvious that again a parabolic temperature behavior is found with a peak at T= TOif

9.880 V (10.000 V is possible). 2.2 mV for 60-K change. <10 ppm/month.

(Vo)ra =

m(l + &)(vO)To.

(1!5)

200/.Lv.,. 70 pV for 5-mA variation.

1.3 KS for 5-mA load variation; 2.2 MS for 10-mA load variation. 13-30 v.

A!

.-

Therefore a peak voltage of 10.000 V at 300 K is obtained with m = 8, if a = 0.012. This was confirmed by measurements.

Supply voltage

226

IEEE JOURNAL

OiT SOLID-ST.kTE CIRCUITS, JUNE

1973 prac-

diode tically

sources,

and

has

the

added

advantage

that

no selection

is necessary. ACKNOWLEDGMENT

Theauthor careful

wisllestothank of nearly

J. H. J. Kokenberg all measurements.

for the

execution

REFERENCES

-b~
-..+d
Fig. 10.

u
Proposed

m d~odepars

R6

-&

method

to obtain

other output

voltages.

supply voltage Outpul current supply current (unloaded)

sensitivity

1 ~V/V. 10 mA 2.3 mA. continuous.

[11 E. Hineman and J. Roberson, Solid state voltage references, Instwm. Cent,. Syst., pp. 133-135,,Mm. 1971, [21 R. P. Baker and J. Nagy, Jr., An investigation of long-term stability of Zener voltage references, IRE i%ans, Instrum,, vol. I-9, pp. 226231, Sept. 1960, [31 D. F. Hilbiber, A new semiconductor voltage standard, in ISSCC Dig. Tech. Papers, 1964, pp. 3233, [41 C. M. Hart and C. Mulder, An all-silicon timing circuit for automatic cameras, MicroeZectron. Ret., vol. 9, pp. 335-340, 1!270 [51 R. J. Widlar~ New developments in IC voltage regulators, IEEE J. So[td-State Cim@,vol. SC-6, pp. 2-7, Feb. 1971. iYew York: [61 S. M. Sze, Ph~sics of Semzcmrductor Deuices. Wiley, 1969, pp. 27, 39, 41, 269,

VI. CO~CLUSIONS The adjustable reference source described, which supplies a voltage of about 10 V, has a parabolic temperature dependence with a drift of about 2.2 mV over 60 K. The circuit has low-noise, good long-term stability, good rejection of supply voltage variations, and a small time constant for load variations. The performance of the circuit can be compared with the best available Zener

Karel E. Kuijk was born in Dcveni,er, The Netherlands, on November 16, 1940, He received the Ir, degree in electrical engineering from the Technical University of Eindhoven, Eindhoven, The Netherlands, in 1964. Since 1965 he has been working in the Philips Research Laboratories, Eindhoven, The Netherlands, on circuit design for instrumentation. He has patents pending in the areas of triangular wave generation and linear integrated circuits.

--b+
x-

.-n-;:

n-Channel Ion-Implanted Enhancement/Depletion FET Circuit and Fabrication Technology

LEONARD FORBES

Abstrac f-n-channel

enhancement/depletion

technolozv -.

and

circuits are described. The threshold voltage adjustment to form the enhancementand depletion-mode devices are achieved by ion implantation. This allows optimization of the performance and circuit density. The calculated arrd experimentally observed speedpower product is 10 pJ/pF with a single +5-V power supply. Inversion of the field region on the high resistivity p-type substrate is completely eliminated by the use of an implanted field shield. INTRODUCTION FET load devices can give a two- to tenfold improvement in the speedD power product. of FET circuits in comparison to linear load resistors or saturated mode FET load devices [1], [2]. A repletion-mode device is normally ON and acts almost like a constant current source charging the EPLETIONT-MODE

v =-i ~1.~;.

Ta.

Manuscript received October 11, 1972; revised January 2, 1973. The author was with the IBM Components Division, Manassas, He is now with the Department of Electrical Engineering, University of Arkansas, Fayetteville, Ark. 72701.

load capacitance during the period of time when the output voltage is rising. The speed-power product improvement is due to the fact that the load capacitance is charged by a constant current source rather than a linear or nonlinear load resistor. This report describes an n-channel FET technology and circuits that employ enhancement-mode switching devices and depletion-mode load devices. The threshold voltage adjustments to form the enhancement- and depletion-mode devices on the same silicon chip are achieved by ion implantation. Ion implantation allows the threshold voltages to the adjusted so as to optimize the performance and circuit density. A single +5-V power supply is used to make the circuits compatible with the standard transistortransistor logic (TTL) bipolar circuits. Field inversion on the high resistivity p-type substrate is completely eliminated by the use of an ionimplanted field shield. The following sections will describe the device design,