- Email: [email protected]
On the time-delay in chalcogenide glass threshold switches
JOURNAL OF NON-CRYSTALLINE SOLIDS 8--10 (1972) 422--426 © North-Holland Publishing Co.
ON THE TIME-DELAY IN CHALCOGENIDE GLASS T H R E S H O L D SWITCHES
S. H. LEE, H. K. HENISCH* and W. D. BURGESS Materials Research Laboratory, The Pennsylvania State University, University Park, Pennsylvania 16802, U.S.A. It is shown that two distinct operating regimes exist, a low voltage regime (close to the threshold point) under which the switching delay is subject to substantial fluctuations and a high voltage regime (high overvoltages) for which the delay is closely determined. The results show that the origin of the scatter lies not primarily in the mechanism whereby the threshold point is approached but in the transient and semi-permanent after-effects of previous switching events. These after-effects can be detected in terms of a diminished pre-threshold conductance.
1. Introduction Threshold switching in a variety of amorphous semiconductors0 and especially in chalcogenide glasses 2) has been extensively studied in recent years. However, the fact that these processes are statistical in character has received very little attention. When switches are repeatedly "addressed" with pulses of constant voltage and substantial duration, the statistical character shows itself as a spread in the observed switching delays; when switches are addressed with voltage pulses of fixed width and adjustable amplitude, it shows itself as a spread in the observed switching voltage V. Fig. 1 shows this and also the (mean) switching delay tD as a function of the applied voltage V. This relationship has been represented 3,4) by the empirical equation tD = tDo exp [-- ( V -- VTH)/Vo],
(1)
where tD0 and Vo are constants, and VTH is the threshold voltage, in principle the minimum voltage but for practical reasons usually the " m o s t probable" switching voltage. For theoretical attempts to deduce eq. (1) see refs. 5 and 6. Since to is the subject of a statistical distribution, it is reasonable to explore, whether this arises from a corresponding fluctuation of VrH. When * Also affiliated with the Department of Physics. 422
ON
TIIE
423
TIME-DELAY
V is large, such fluctuations are obviously unimportant, but the overvoltage concept V-VTH c e a s e s to have a precise meaning when V and VTn are compatible. All the results here discussed were obtained on encapsulated thin (-,~ 1 lam) chalcogenide glass films between polished graphite electrodes*. Starting ma-
I
tD----e
(o)
o. Time
N
tD
(b) , u
OFF
v
l
Ib Time
15
(c)
o
~
p.
13 I
I
I
2
3
4
T i m e Del0y (p.sec)
Fig. 1. Schematicrepresentation of statistical threshold switching aspects. (a) constant voltage; spread of delay times, V~ VTR. (b) constant delay time; spread of switching voltages. (c) relationship between applied voltage and switching time tD.
terial: Te4oSiisGevAsa5. The properties of such switches (except for aspects connected with working life) do not depend at all sensitively on composition, and the results should therefore have a more general validity than a single composition would suggest. * Obtained from Energy Conversion Devices, Inc., through the courtesy of Mr. S. R. Ovshinsky.
424
S. H. LEE, H. K. H E N I S C H A N D W . D. BURGESS
2. Experimental results
Fig. 2a shows a series o f switching time delays associated with a h u n d r e d events at each a p p l i e d voltage. I f eq. (1) were the only relevant relationship a n d VTH the only fluctuating quality, then it w o u l d follow t h a t the u n c e r t a i n t y At o w o u l d be directly p r o p o r t i o n a l to to, b u t fig. 2a shows t h a t this is n o t the
tD
I0
J
t
I
I
I
(a)
w
0.5 0.3 13
i
to
"G
14
f
15 16 17 Voltage (V)
i
i
[ON = I 5 mA
I
i
o°
18
i
19
i
i
i
room temperature
6 % %
5
o %
(b)
%
o% o
~3
o o%
°°
%~'
o° ION = 6 m A
"
"".,,
oN oeeOo% I
IO
I
I
20
I
I
30
I
I
°%oOWe% I
40
t
I
50
No. of Switching Event
Fig. 2. Characteristics of statistical switching effects. (a) distinct high and low voltage regimes. VTn = 13.5 V at 293°K and 16.5 V at 210°K. Dots denote mean values. (b) effect of on-current on statistical behavior. Results up to broken line correspond to constant over-voltage within 0.2 V. (30 sec intervals; 6 lasec pulses N 0.5 V above VTrt.)
case. T h e r e is no statistical regime at all for small values o f t D. A s the overvoltage diminishes, the onset o f scatter is very sudden, m u c h m o r e so t h a n is generally believed. F o r the case u n d e r investigation, V0 ~ 2 V, A VxH < 0 . 2 V a n d the longest delay t o = 4 psec. This w o u l d give Ato = 0 . 4 psec whereas the o b s e r v e d value is a b o u t 2 psec. The width o f the statistical regime decreases
ON THE TIME-DELAY
425
towards higher-temperatures. Its general form follows
Nto/No = exp ( - tD/to),
(2)
except at high values of tD where NtD/No diminishes sharply. For IoN < 6 mA, AtD and tD remains constant during a long series of successive switching events. 3. Discussion
At room temperature, t o commonly fluctuates by a factor of 2, and at low temperatures by as much as 4, as shown in fig. 2a. Since the threshold current is almost constant, this implies fluctuations of input energy by the same factor, and corresponding variations of the maximum temperature reached in the course of self-heating. If thermal filament formation due to self-heating were the essential cause of switching, there would seem to be no reason for such a pronounced random fluctuation of the energy requirements. Near V ~ VTn the switching conditions (possibly the length, condition or location of the ON-state filament in the course of its formation) are not closely defined. Increasing over-voltages imply ultimately an increasing local temperature, even if the switching mechanism at V~ VTn is essentially nonthermal, and thereby a more precise localization of the switching event and a corresponding reduction of AtD. However, there is no reason for believing that this process should set in as abruptly as fig. 2a shows (considering the logarithmic scale). Thus, although heating may play some localizing role, it is more plausible to conclude that two distinct electronic processes are at work, competing with one another at low voltages. At high voltages the process responsible for the fluctuations evidently disappears. Before a detailed mechanism can be examined, it is essential to know whether the fluctuations are a peculiarity of the excitation process (i.e. the mechanism whereby the threshold condition is attained) or are variable aftereffects of the previous switching event. It has been customary to assume the former, but recent experiments have demonstrated that the fluctuations arise in fact, wholly or partially, from after-effects. It can be shown 10) that the prethreshold resistance of a switch is reduced by a previous switching event. In the first microsecond thereafter (at room temperature), the resistance increases by 2 or 3 orders of magnitude, during the next second by about 3½ orders. It then continues to increase detectably for several minutes until a static value is reached. Since these changes take place after the cessation of the filament current, it is plausible to conclude that they involve not only the filament region but the entire electrode area. The final value fluctuates from event to event by 8-9 percent. A threshold switch is evidently in a different "state" after each operation, either because of a re-distribution of
426
S . H . LEE, H . K . H E N I S C H AND W . D . BURGESS
residual charge or because of some minor structural change. In either case, it would have to be assumed that the effect is obliterated when a sufficiently high over-voltage is applied, as a minor internal field perturbation might be. The random nature of the fluctuations would seem to favor a nonstructural interpretation, e.g. possibly one dependent on carrier trapping in association with the compositional heterogeneity of the materials11,12). The decay of the after-effects is electric-field dependent in a manner which is qualitatively consistent with this hypothesis.
Acknowledgement This research was supported by the Advanced Research Projects Agency of the Department of Defense and was monitored by G. Boghosian, U.S. Army Research Office, Durham, N.C., under contract No. D A H C0470C 0047. The authors also wish to thank Sir Nevill Mott, S. R. Ovshinsky, R. W. Pryor and R. Shaw for valuable discussions.
References 1) 2) 3) 4) 5) 6) 7) 8) 9) 10) 1l)
J. G. Simmons, Contemp. Phys. U (1970) 21. See for instance, J. Non-Crystalline Solids 2 (1970) and 4 (1970). R. R. Shanks, J. Non-Crystalline Solids 2 (1970) 504. S. R. Ovshinsky, Phys. Rev. Letters 21 (1970) 1450. F. Schmidlin, Phys. Rev. 1 (1970) 1583. K. W. Bo6r, J. Appl. Phys. 41 (1971) 2675. H. K. Henisch, E. A. Fagen and S. R. Ovshinsky, J. Non-Crystalline Solids 4 (1970) 538. H. K. Henisch and R. W. Pryor, Solid State Electron. 14 (1971) 765. S. H. Lee and H. K. Henisch, to be punished. W. Heywang and D. R. Haberland, Solid State Electron. 13 (1970) 1077. H. Fritzsche, J. Non-Crystalline Solids 6 (1971) 49.
ON THE TIME-DELAY IN CHALCOGENIDE GLASS T H R E S H O L D SWITCHES
S. H. LEE, H. K. HENISCH* and W. D. BURGESS Materials Research Laboratory, The Pennsylvania State University, University Park, Pennsylvania 16802, U.S.A. It is shown that two distinct operating regimes exist, a low voltage regime (close to the threshold point) under which the switching delay is subject to substantial fluctuations and a high voltage regime (high overvoltages) for which the delay is closely determined. The results show that the origin of the scatter lies not primarily in the mechanism whereby the threshold point is approached but in the transient and semi-permanent after-effects of previous switching events. These after-effects can be detected in terms of a diminished pre-threshold conductance.
1. Introduction Threshold switching in a variety of amorphous semiconductors0 and especially in chalcogenide glasses 2) has been extensively studied in recent years. However, the fact that these processes are statistical in character has received very little attention. When switches are repeatedly "addressed" with pulses of constant voltage and substantial duration, the statistical character shows itself as a spread in the observed switching delays; when switches are addressed with voltage pulses of fixed width and adjustable amplitude, it shows itself as a spread in the observed switching voltage V. Fig. 1 shows this and also the (mean) switching delay tD as a function of the applied voltage V. This relationship has been represented 3,4) by the empirical equation tD = tDo exp [-- ( V -- VTH)/Vo],
(1)
where tD0 and Vo are constants, and VTH is the threshold voltage, in principle the minimum voltage but for practical reasons usually the " m o s t probable" switching voltage. For theoretical attempts to deduce eq. (1) see refs. 5 and 6. Since to is the subject of a statistical distribution, it is reasonable to explore, whether this arises from a corresponding fluctuation of VrH. When * Also affiliated with the Department of Physics. 422
ON
TIIE
423
TIME-DELAY
V is large, such fluctuations are obviously unimportant, but the overvoltage concept V-VTH c e a s e s to have a precise meaning when V and VTn are compatible. All the results here discussed were obtained on encapsulated thin (-,~ 1 lam) chalcogenide glass films between polished graphite electrodes*. Starting ma-
I
tD----e
(o)
o. Time
N
tD
(b) , u
OFF
v
l
Ib Time
15
(c)
o
~
p.
13 I
I
I
2
3
4
T i m e Del0y (p.sec)
Fig. 1. Schematicrepresentation of statistical threshold switching aspects. (a) constant voltage; spread of delay times, V~ VTR. (b) constant delay time; spread of switching voltages. (c) relationship between applied voltage and switching time tD.
terial: Te4oSiisGevAsa5. The properties of such switches (except for aspects connected with working life) do not depend at all sensitively on composition, and the results should therefore have a more general validity than a single composition would suggest. * Obtained from Energy Conversion Devices, Inc., through the courtesy of Mr. S. R. Ovshinsky.
424
S. H. LEE, H. K. H E N I S C H A N D W . D. BURGESS
2. Experimental results
Fig. 2a shows a series o f switching time delays associated with a h u n d r e d events at each a p p l i e d voltage. I f eq. (1) were the only relevant relationship a n d VTH the only fluctuating quality, then it w o u l d follow t h a t the u n c e r t a i n t y At o w o u l d be directly p r o p o r t i o n a l to to, b u t fig. 2a shows t h a t this is n o t the
tD
I0
J
t
I
I
I
(a)
w
0.5 0.3 13
i
to
"G
14
f
15 16 17 Voltage (V)
i
i
[ON = I 5 mA
I
i
o°
18
i
19
i
i
i
room temperature
6 % %
5
o %
(b)
%
o% o
~3
o o%
°°
%~'
o° ION = 6 m A
"
"".,,
oN oeeOo% I
IO
I
I
20
I
I
30
I
I
°%oOWe% I
40
t
I
50
No. of Switching Event
Fig. 2. Characteristics of statistical switching effects. (a) distinct high and low voltage regimes. VTn = 13.5 V at 293°K and 16.5 V at 210°K. Dots denote mean values. (b) effect of on-current on statistical behavior. Results up to broken line correspond to constant over-voltage within 0.2 V. (30 sec intervals; 6 lasec pulses N 0.5 V above VTrt.)
case. T h e r e is no statistical regime at all for small values o f t D. A s the overvoltage diminishes, the onset o f scatter is very sudden, m u c h m o r e so t h a n is generally believed. F o r the case u n d e r investigation, V0 ~ 2 V, A VxH < 0 . 2 V a n d the longest delay t o = 4 psec. This w o u l d give Ato = 0 . 4 psec whereas the o b s e r v e d value is a b o u t 2 psec. The width o f the statistical regime decreases
ON THE TIME-DELAY
425
towards higher-temperatures. Its general form follows
Nto/No = exp ( - tD/to),
(2)
except at high values of tD where NtD/No diminishes sharply. For IoN < 6 mA, AtD and tD remains constant during a long series of successive switching events. 3. Discussion
At room temperature, t o commonly fluctuates by a factor of 2, and at low temperatures by as much as 4, as shown in fig. 2a. Since the threshold current is almost constant, this implies fluctuations of input energy by the same factor, and corresponding variations of the maximum temperature reached in the course of self-heating. If thermal filament formation due to self-heating were the essential cause of switching, there would seem to be no reason for such a pronounced random fluctuation of the energy requirements. Near V ~ VTn the switching conditions (possibly the length, condition or location of the ON-state filament in the course of its formation) are not closely defined. Increasing over-voltages imply ultimately an increasing local temperature, even if the switching mechanism at V~ VTn is essentially nonthermal, and thereby a more precise localization of the switching event and a corresponding reduction of AtD. However, there is no reason for believing that this process should set in as abruptly as fig. 2a shows (considering the logarithmic scale). Thus, although heating may play some localizing role, it is more plausible to conclude that two distinct electronic processes are at work, competing with one another at low voltages. At high voltages the process responsible for the fluctuations evidently disappears. Before a detailed mechanism can be examined, it is essential to know whether the fluctuations are a peculiarity of the excitation process (i.e. the mechanism whereby the threshold condition is attained) or are variable aftereffects of the previous switching event. It has been customary to assume the former, but recent experiments have demonstrated that the fluctuations arise in fact, wholly or partially, from after-effects. It can be shown 10) that the prethreshold resistance of a switch is reduced by a previous switching event. In the first microsecond thereafter (at room temperature), the resistance increases by 2 or 3 orders of magnitude, during the next second by about 3½ orders. It then continues to increase detectably for several minutes until a static value is reached. Since these changes take place after the cessation of the filament current, it is plausible to conclude that they involve not only the filament region but the entire electrode area. The final value fluctuates from event to event by 8-9 percent. A threshold switch is evidently in a different "state" after each operation, either because of a re-distribution of
426
S . H . LEE, H . K . H E N I S C H AND W . D . BURGESS
residual charge or because of some minor structural change. In either case, it would have to be assumed that the effect is obliterated when a sufficiently high over-voltage is applied, as a minor internal field perturbation might be. The random nature of the fluctuations would seem to favor a nonstructural interpretation, e.g. possibly one dependent on carrier trapping in association with the compositional heterogeneity of the materials11,12). The decay of the after-effects is electric-field dependent in a manner which is qualitatively consistent with this hypothesis.
Acknowledgement This research was supported by the Advanced Research Projects Agency of the Department of Defense and was monitored by G. Boghosian, U.S. Army Research Office, Durham, N.C., under contract No. D A H C0470C 0047. The authors also wish to thank Sir Nevill Mott, S. R. Ovshinsky, R. W. Pryor and R. Shaw for valuable discussions.
References 1) 2) 3) 4) 5) 6) 7) 8) 9) 10) 1l)
J. G. Simmons, Contemp. Phys. U (1970) 21. See for instance, J. Non-Crystalline Solids 2 (1970) and 4 (1970). R. R. Shanks, J. Non-Crystalline Solids 2 (1970) 504. S. R. Ovshinsky, Phys. Rev. Letters 21 (1970) 1450. F. Schmidlin, Phys. Rev. 1 (1970) 1583. K. W. Bo6r, J. Appl. Phys. 41 (1971) 2675. H. K. Henisch, E. A. Fagen and S. R. Ovshinsky, J. Non-Crystalline Solids 4 (1970) 538. H. K. Henisch and R. W. Pryor, Solid State Electron. 14 (1971) 765. S. H. Lee and H. K. Henisch, to be punished. W. Heywang and D. R. Haberland, Solid State Electron. 13 (1970) 1077. H. Fritzsche, J. Non-Crystalline Solids 6 (1971) 49.