 |
Description  |
|
|
CROSS-REFERENCE TO RELATED APPLICATIONS
The following copending applications disclosed related subject matter: Ser.
Nos. 626,802, 626,808, 626,809, 626,342, 626,806, and 626,807, all filed
June 29, 1984.
BACKGROUND AND SUMMARY OF THE INVENTION
The present invention relates to electronic devices.
It is generally recognized that conventional VLSI integrated technical
technology will be prevented from further scaling by the time MOS devices
get down to a quarter micron channel length, and perhaps even at much
larger geometries. Since much of the advance in integrated circuit
capabilities has been based on the continued progress of scaling, this
near-future barrier is of substantial concern.
Thus it is an object of the present invention to provide an integrated
circuit technology wherein active devices can have active regions smaller
than one quarter micron in dimension.
It is further object of the present invention to provide an integrated
circuit technology wherein active devices can be fabricated which occupy a
total area of less than 1/4 of a square micron average for each active
device.
A further inherent limitation of conventional integrated circuit technology
is speed. MOS devices have inherent limits on their speed due to the
channel-length transit time. Intergrable bipolar devices also have
inherent speed limitations, due to the base width transit time, and are
also likely to have high power dissipation.
Thus it is an object of the present invention to provide an active device
having higher potential maximum speed than any MOS device.
It is a further object of the present invention to provide an active device
which is potentially faster than any bipolar device.
It is a further object of the present invention to provide an active device
which is potentially faster than any bipolar device and which also has a
very low power dissipation.
To achieve these and other objects, the present invention provides: a new
genus of electronic devices, wherein at least two closely adjacent
potential wells (e.g. islands of GaAs in an AlGaAs lattice) are made small
enough that at least two components of momentum of carriers within the
wells are discretely quantized. This means that, when the bias between the
wells is adjusted to align energy levels of the two wells, tunneling will
occur very rapidly, whereas when energy levels are not aligned, tunneling
will be greatly reduced. This high-gain mechanism leads to useful
electronic device functions.
A difficulty in making quantum-coupled devices into functional electronic
circuits is that these devices are so extremely small that it is typically
necessary to run a number of them in parallel to provide macroscopic
output currents. In addition, the routing of wiring to couple into and out
of these multiple parallel active devices is also difficult, since the
tight geometry constraints of the devices place substantial constraints on
the geometry which must be used for the wiring.
Thus, it an object of the present invention to provide a device structure
using quantum-coupled devices wherein input and output connections to
route macroscopic currents to and from the devices are provided.
According to the present invention there is provided:
An electronic device comprising:
a plurality of first and second potential wells each comprising an island
of a semiconducting material having a minimum dimension less than 500
Angstroms and another dimension less than 1000 Angstroms;
a barrier medium interposed between said first and second wells wherein the
minimum potential energy of carriers is at least 50 millielectronvolts
higher than the minimum potential energy of carriers within said wells,
said wells being physically separated by a distance which is less than
three times the smallest physical dimension of either of said wells;
a first conductor electrically coupled to each of said first wells;
a second conductor electrically coupled to each of said second wells; and
an output contact laterally separated from plural ones of said second wells
by said barrier medium, said barrier medium between said output contact
and said second wells having a minimum lateral width which is no more than
twice the minimum lateral width separating said first and second wells;
wherein each of said wells comprises a lightly doped semiconductor, and
each said conductor comprises a heavily doped semiconductor.
BRIEF DESCRIPTION OF THE DRAWINGS
The present invention will be described with reference to the accompanying
drawings, wherein:
FIGS. 1A and 1B shown spacing and energy levels of potential wells in a
simple sample embodiment;
FIG. 2 shows the structure of FIG. 1, biassed to permit resonant tunneling;
FIG. 3 shows the structure of FIG. 1, biassed at a lower voltage than shown
in FIG. 2, so that resonant tunneling is forbidden;
FIG. 4 shows voltage-current characteristics of the structure of FIG. 1;
FIG. 5 shows the energy levels of an embodiment using moderately large
wells, wherein the upper-lying energy levels are more closely spaced;
FIG. 6 shows the electronic structure of an embodiment including input and
output contacts;
FIG. 7 shows the energy levels of an embodiment using approximately
sinusoidal well boundaries, wherein the energy levels are more equally
spaced;
FIG. 8 shows a three-terminal quantum-well device according to the present
invention;
FIG. 9 shows another three-terminal quantum-well device according to the
present invention, wherein multiple chains of quantum-well pairs are
connected in parallel;
FIG. 10 shows a cross-section of the device of FIG. 9;
FIG. 11 shows a many-terminal device, which differs from the device of FIG.
9 in having multiple electrodes 208 prime for connection to multiple wells
204 prime in each chain;
FIG. 12 shows a sample set of bias conditions for the embodiment of FIG. 11
wherein resonant tunneling occurs, and FIG. 13 shows a sample set of bias
conditions for the embodiment of FIG. 11 wherein resonant tunneling does
not occur;
FIG. 14 shows a read-only memory according to one embodiment of the
invention;
FIG. 15 shows a sample output switch configuration used in practicing the
present invention, wherein quantum-well devices switch macroscopic output
currents;
FIG. 16 shows the effect of the electric potential change induced by the
population of well levels on the permissible tunneling transitions in a
further class of embodiments;
FIG. 17 shows an AND gate configured using quantum-well logic and
self-consistent tunneling restrictions, and FIG. 18 shows a more complex
logic element configured using quantum-well logic with self-consistent
tunneling restrictions;
FIGS. 19 and 20 show two embodiments of vertical-tunneling quantum-well
device structures according to the invention; and
FIGS. 21-36 show steps in processing a sample three-terminal device
according to the present invention.
DESCRIPTION OF THE PREFERRED EMBODIMENTS
The present invention teaches active devices which are fundamentally
different from the transistor and diode structures which have been used
heretofore in the electronics industry to configure integrated circuits.
That is, active devices taught by the present invention are neither field
effect transistors nor bipolar transistors, but operate according to
principles which are fundamentally and entirely different.
A key feature of many embodiments of the invention is a pair of closely
coupled quantum wells which are each extremely small, so small that their
electron populations are quantized, i.e. an electron in a well can only
have one of a few allowable energies. The present invention teaches modes
of device operation which cannot be analyzed by classical physics, but
must be understood in light of quantum mechanics.
Advances in semiconductor processing now permit semiconductor structures to
be patterned with exceedingly small dimensions, comparable to the Bloch
wave length of an electron. (The Bloch wavelength measures the "width" of
an electron in a semiconductor. The location of an electron must be
measured by a probability distribution, which has a certain width.) The
present invention makes use of the availability of patterned structures in
dimensions which are close to the Bloch wave length to achieve new types
of device structures, capable of new principles of operation of a kind not
seen before in semiconductor or integrated circuit devices.
A simplified sample embodiment, which demonstrates some of the key
principles of operation used in the present invention, is shown in FIG. 1.
Separate wells of gallium arsenide are embedded in a matrix of aluminum
gallium arsenide. As is well known in the semiconductor art, the bandgap
of AlGaAs is wider than that of GaAs. (More precisely, the pseudo-binary
alloys of the formula Al.sub.x Ga.sub.1-x As are all semiconductors, with
a bandgap which increases as a function of x.) This different bandgap
means that conduction-band electrons in the lattice see a lower potential
energy in the GaAs regions than in the AlGaAs regions. A particular
advantage of this material system is that the interface between GaAs and
AlGaAs is a very well behaved one, that is such interfaces can be
fabricated with an extremely low density of surface states (less than the
Si/SiO.sub.2 interface), and can preserve lattice match almost perfectly
between the GaAs and the AlGaAs sides of the interface. However, as will
be discussed below, the present invention can be practiced with a wide
variety of material systems, and not only semiconductors.
In the sample embodiment of FIG. 1, the wells 10 are about 125 angstroms
wide, and the spacing between the wells is also about 125 angstroms. The
dimensions of these wells are small enough that the energy states of
electrons in the well will be quantized. That is, these wells are
preferably made cubic, and the allowable energy states of an electron in a
cubic well of such dimensions can be calculated easily, using the
effective mass approximation for the behavior of the electron, as a simple
particle in a box problem in quantum mechanics. Thus, for example, for the
well dimensions given above, and an aluminum concentration of 35%, there
exist four allowable energy states, and the lowest lying energy state will
be above the bottom of the GaAs conduction band minimum and will be
separated from the next lowest energy state by 0.070 electron volts. The
next higher lying energy state will lie another 0.114 electron volts
higher. Note that these energy states must be treated as quantized. That
means that the energy of an electron in the well must be exactly equal to
one of these permitted discrete energy levels and cannot take any other
value. This is key to the present invention. Note that this is a
fundamental difference from normal semiconductor devices (even from the
advanced semiconductor devices projected to be achievable by scaling
present semiconductor devices), in that all conventional solid state
devices--by they field effect or bipolar transistors or anything
else--require that the energy levels of an electron within the conduction
band of the semiconductor be quasi-continuous.
It should be noted that the number of electrons which can populate each
energy level in such a quantized system is strictly limited. That is, the
lowest lying energy level in each well in this example can be populated by
only two electrons, the next level up can be populated by only six
electrons, the next level up can be populated by only ten electrons, and
so on. This means that the number of carriers within a well is strictly
limited. Thus, a carrier cannot tunnel to an energy state in the next well
if that energy state is already completely filled. While each well can
hold only a few carriers at a time, it should be noted that the transition
times are expressly short (typically far less than a pico-second), and
total current can be increased by operating many pairs of wells in
parallel.
Now, in the sample embodiment the wells are placed sufficiently close such
that the probability density of an electron in one well extends into the
neighboring well; that is, the wells 10 and 12 in FIG. 1 are close enough
together that electrons can tunnel between them easily. In particular, as
shown in FIG. 1b, an electron in the first well can tunnel easily to a
state of equal energy in the second well, but of course it cannot do this
unless an unoccupied state having the same energy actually exists in the
second well. However, suppose that the energy levels in the two wells do
not line up. Then an electron in the first well can only tunnel to states
of the second well which have a different energy than the first well, and
it can only tunnel to such states if there is also some mechanism whereby
the electron can lose or gain enough energy to reach the energy of the
lower-lying available state in the second well.
The chief mechanism for the changing of the energy of an electron to
achieve tunneling in this fashion is relaxation via surface states. That
is, even the GaAs to AlGaAs interface, which is exceedingly benign as
semiconductor interfaces go, will have a measurable density of surface
states at the interface. Such surface states provide a scattering process
which will permit an electron to transfer to higher or lower energy, if it
intersects one of the available sites for scattering. That is, the
interface can be described as having an areal density of scattering sites
which are capable of changing the energy from an electron so that it can
make the transition between wells even when the energy levels of the wells
do not line up. This mode of transition is known as inelastic tunneling.
The alternative mode of tunneling, wherein the electron tunnels from the
first well to the second with no change in energy, is known as resonant
tunneling.
Tunneling requires conservation both of energy and also of momentum.
Momentum is also quantized for small potential wells. That is, in each
dimension of the box which defines the boundaries of the potential well,
the smaller the length of that dimension the more widely spaced the
allowable values of that component of momentum will be. Thus, as reported
in the Soloner et al. article cited below, resonant tunneling can occur
between two very closely spaced thin sheets of lower potential. Since the
potential wells are very large in two dimensions, the momentum components
in those two dimensions are not quantized. That is, when one dimension of
a potential well is large, the momentum components in that direction are
so closely spaced that a background lattice phonon will be available to
supply the momentum difference. That is, the density of phonons found in
the lattice follows Bose-Einstein statics, i.e. the number of phonons
found in the lattice at energy E in a background lattice temperature T
varies as 1 over (EXP ((E/k T)-1)). This energy distribution of phonons
also implys a distribution for the magnitude of momentum of the phonons,
and, since the phonons can be treated as approximately isotropic in
reasonable semiconductor materials, this same distribution also specifies
the distribution of phonons having a desired value of some one particular
component of momentum. Thus, the quantization of momentum in a potential
well can be considered as discrete or quasi-continuous only in relation to
temperature. That is, for example, a 1000 angstrom wide box will result in
momentum quantization at one degree Kelvin, since the phonon population
will be crowded into energetically low states, but at 300.degree. K. there
would be a tremendous lattice background phonon population to bridge gaps
between the allowable momentum values, and thus the separation between the
allowable momentum values in this direction would not be meaningful, i.e.
momentum in this direction would have to be treated as a quasi-continum
value, i.e. any value of this momentum component which an incoming carrier
had would still permit tunneling, since lattice phonons could adjust this
momentum component to one of the allowable values. If all three components
of momentum are discretely quantized, then it follows that energy must be
discretely quantized also. The quantization of energy levels is (to a
rough approximation) merely dependent on the volume of the potential well,
so that a thin flat potential well could have quantized energy states and
sharply separate discrete allowable levels for one component of momentum,
whereas the allowable values of the other two momentum components will be
closely spaced, so that they impose no practical constraint on tunneling.
That is, the resonant tunneling gain is enhanced in accordance with the
number of parameters which must be met. In tunneling between two closely
spaced cubic potential wells, all three components of momentum (and
therefore energy) must match between the state which the carrier is
leaving in the first well and the state which the carrier is entering in
the second will. That is, if an incoming carrier does not satisfy all
three momentum constraints, the chances of it finding a lattice phonon
which has exactly the right components to satisfy the difference in each
of the three momentum values is quite small. However, if only one
component must be adjusted, the chances of accidentally doing this by
interaction of the lattice phonon are much greater.
Thus, the preferred embodiments of the present invention use structures
wherein all three components of momentum are quantized. However, a class
of less preferred alternative embodiments of the present invention uses
structures wherein only two components of momentum but not energy are
quantized, i.e. closely spaced thin wires. It is easy to satisfy the
momentum constraint alone by a phonon-assisted process, but the necessity
for satisfying both energy and momentum conversation is what keeps the
rate for inelastic tunneling background controlled. That is, if a
scattering center at the metallurgical interface can change the energy of
a carrier, it is easy for one of the lattice phonons to provide the
appropriate adjustment in momentum.
As device dimensions are made smaller, two desirable effects are achieved:
First, the separation of the quantized energy states in a well increases
as the well dimension becomes smaller. Second, as the wells become closer
together, the rate of tunneling is increased. That is, the probability of
tunneling can be expressed as a constant times exp(-2d.times.f(E)), where
d is the distance between the wells and E is the energy difference.
This exponential dependance means that the probability of resonant
tunneling is tremendously increased as the distance between wells is
reduced. Inelastic tunneling will not be increased comparably, since as
noted above inelastic tunneling is limited, in good material, by the
density of scattering centers. Thus, room temperature operation becomes
possible at interwell spacing of about 125A or less.
A limitation on resonant tunneling is provided by thermal "smearing", i.e.
redistribution of energy level populations caused by background thermal
energy. That is, the density of states distribution within each well of
the discrete energy levels is broadened somewhat by thermal smearing. At
higher temperatures, the thermal smearing is greater. The problem thermal
smearing causes is that the operating temperature must be low enough that
thermal smearing does not populate higher energy level states. That is,
the electrons should reside in the lowest energy level to eliminate
accidental coincidences and bidirectional elastic tunneling. This reduces
to the condition that the energy spacing must be much larger than the
thermal energy of the electrons. Thus, for operation at 4.degree. K. with
GaAs wells in an Al.sub.0.3 Ga.sub.0.7 As matrix, the well width (and well
spacing) should be in the neighborhood of 0.1 to 0.2 microns or less.
However, to increase the operating temperature to 300.degree. K., the
critical dimension needs to be reduced merely to a number on the order of
125 Angstroms or less.
It should be further noted that the energy levels in adjacent wells can be
made to line up or not line up simply by applying a bias, as shown in FIG.
2. In this case, a sufficient bias voltage has been applied in the
direction of conduction, so that the second level in the second well is
lined up with the first level in the first well. Under these conditions
resonant tunneling will regularly occur, and the fast relaxation from the
second level to the ground state in the second well ensures that the
device is unidirectional.
If zero bias were applied to this same structure, resonant tunneling would
still be possible, but would be bidirectional. That is, at zero bias the
energy levels in two adjacent wells will line up, but electrons will
tunnel from the first well to the second well just as fast as they tunnel
from the second well to the first well. Note, however, that if half as
much bias as shown in FIG. 2 is applied, resonant tunneling will be
forbidden (at sufficiently low temperature) and only inelastic tunneling
is permissible. In this bias condition, as shown in FIG. 3, an electron
can make a transition from a state in well one to an lower lying state in
well two only if a scattering event also occurs, as discussed above. Thus,
at small dimensions, the frequency of inelastic transitions is limited by
the density of available scattering sites. However, the density of
scattering sites is in effect an areal density, since it arises primarily
from surface states rather than from bulk defects or particles within the
crystal lattice. Thus, since the limiting factor is a areal density of
defects, the inelastic tunneling current is relatively insensitive to the
distance between wells. This means that, for good quality material, as the
dimensions are scaled the resonant tunneling current increases
tremendously, but the inelastic tunneling current does not increase much.
The result of these phenomena is that the current/voltage graph for
tunneling between two wells looks approximately as seen in FIG. 4. That
is, inelastic tunneling will supply a current which increases
approximately exponentially with the applied voltage. Resonant tunneling
will also supply at least one current peak added on to this curve, at a
bias voltage where the energy levels of the wells line up. Thus, a regime
of substantial negative differential resistance is available, i.e. a
millimeter diode with gain is available, as reported in Sollner et al.,
"Resonant Tunneling Through Quantum Wells at Frequency up to 2.5 THz," 43
Applied Physics Letters 588 (1983). (However, this article used potential
wells in which only one momentum component was discrete as discussed
above.) Other background reference which set forth generally known physics
of resonant tunneling are Chang et al., "Resonant Tunneling in
Semiconductor Double Barriers," 24 Applied Physics Letters 593 (1974); Tsu
et al., "Nonlinear Optical Response of Conduction Electrons in a Super
Lattice," 19 Applied Physics Letters 246 (1971); Lebwohl et al.,
"Electrical Transport Properties in a Super Lattice," 41 Journal of
Applied Physics 2664 (1970); and Vojak et al., "Low-temperature operation
of Multiple Quantum-Well Al.sub.x Ga.sub.1-x As-GaAs P-n Hetero Structure
Lasers Grown by a Metal Organic Chemical Vapor Diposition," 50 Journal of
Applied Physics 5830 (1979). These five references are hereby incorporated
by reference. (It should be noted that portions of the foregoing
discussion merely reflect generally known physics, as reflected in these
articles and pressumably elsewhere, but other portions of the foregoing
discussion are not generally known and do not reflect any understanding
available in the prior art.)
However, the prior art has taught potential wells which are quantized only
in one dimension, which implies relaxed constraints on the selection rules
for elastic tunneling. The preferred embodiments of the present invention
have more stringent selection rules, and should (because all 3 dimensions
are quantized) have dramatically more gain than the 1-D quantized devices.
This invention is significantly different from the Esaki tunnel diode or
2-dimensional electron gas structures in that all three dimensions are
quantized here, which imposes a constraint on the tunneling conditions
much different than other structures; that is, the momentum constraint
significantly depresses other than the desired elastic resonant tunneling.
The foregoing has supposed that only a few quantum levels are found in a
well, but this is not correct for the most convenient well dimensions. A
more realistic energy diagram of the levels in such a structure wherein
1000 angstrom GaAs wells are incorporated in an AlGaAs lattice would be as
shown in FIG. 5. Note that, in the familiar statics of the solution to the
particle in a box problem, the successively higher energy levels are
spaced progressively closer together. This means that, even when the bias
voltage is such that the lowest few levels between the two wells do not
line up, some of the higher lying levels are likely to. However, this is
not a major problem, since the higher lying levels will typically be
depopulated because of the temperature constraint mentioned previously.
That is, the lifetime for relaxation from a higher lying state to an empty
lower state will typically be much shorter than the typical time for
tunneling. Thus, after a lower level has been depleted by tunneling, a
higher level which has been excited in any fashion will tend to relax to
the depleted lower level much faster than it will tunnel. However, this
assumes that empty lower levels exist below the higher level from which
tunneling is possible. This imposes a constraint on the input contact,
which will now be discussed.
FIG. 6 shows the energy diagrams for a system of two coupled wells 10 and
12 plus input and output contacts 14 and 16. Note that the input and
output contacts are doped, although the wells and the areas between them
are preferably not. Note also that two important limitations on the well
size arise from the necessities of making input contact and output
contact. First, the input contact will fill all levels of the first well
10 by tunneling, up to an energy approximately equal to the Fermi level of
the electrons in the doped semiconductor which provides the input contact.
This means that none of the levels thus still filled up must line up with
levels in the second well when tunneling is not desired. That is, if the
wells are too large, they will have discrete energy states at the bottom
of the well but will include densely spaced states below the Fermi level
of the input contact. This means that these densely spaced states will be
filled, and therefore resonant tunneling current will be seen if any of
these dense states line up with states in the second well. This means that
the current gain of the device will be greatly reduced. Second, all levels
in the second well which are above the Fermi level of the output contact
will equilibrate with the output contact, i.e. will essentially always
remain filled. This is not itself a problem, since electrons can tunnel
from the higher lying energy states in the second well into the continuum
of states which exists in the (large) output contact, but if any level in
the first well lines up with one of these filled levels in the second
well, that level in the first well will also remain filled. Note that the
input and output contacts are preferably both degenerately doped n-type.
The physical well shapes have been described as having perfectly sharp
boundaries, but this is not quite realistic. That is, in good quality MBE
material, the transitions will typically be smooth enough that the
potential profile looks more precisely like that drawn in FIG. 7. This is
actually advantageous, since the energy levels will tend to be more nearly
equal spaced. That is, where the energy level spacings are more nearly
equal, there is a wider range of bias conditions wherein a large number of
the lower lying energy levels will not line up.
Of course, for the ultimately preferable small well dimensions, e.g. 125
angstroms, the number of states in each well will be small, e.g. 4 states
per well. With such a small number of states, there will be bias
conditions under which no actual line ups exist.
Each tunneling transition described above is a transition of only a limited
number of electrons for each pair of coincidental states. That is, each of
the discrete energy levels in the quantum well can be populated by only a
certain definite number of electrons. In the example discussed above,
wherein the energy well has the approximate physical shape of a cube, the
lowest lying level can be populated by only two electrons, and no more.
The next higher energy level can be populated by only six electrons, and
no more. The third energy level can be populated by only 12 electrons, and
no more. Most of the higher-lying levels can also be populated by only 12,
although some may have a higher maximum occupancy because of accidental
degeneracies. Thus, it is important to note that a carrier cannot tunnel
into even an allowable energy level unless that allowable energy level is
not completely populated. The multiple states at each allowable energy
level are distinguished by other quantum numbers. That is, for example,
the six electrons in a second energy level can have one of two possible
states of spin, and can have one of three possible momentum vector
directions. However, this distinction among isoenergetic states is
relatively unimportant for understanding the present invention.
Thus, each tunneling transition can transport up to a dozen carriers for
each pair of lined-up wells. More than one pair of wells may be lined up
at the same time. Moreover, many pairs of wells may be operated in
parallel. Moreover, the transit time for tunnelling can be extremely
small, less than a pico second. Thus, although only a few electrons are
translated in each tunneling event, a reasonable current density can be
achieved nevertheless.
Thus, a basic family of embodiments is structures, as described above,
wherein two quantized wells separate an input contact from an output
contact. However, further aspect of the present invention provide numerous
other kinds of innovative device structures.
For example, a three terminal device according to the present invention is
shown in a plan view in FIG. 8. A first quantum dot 202 is coupled to a
second quantum dot 204, which is coupled to an output contact 210. The
dimensions of the quantum dots 202 and 204 are selected as discussed above
for quantum wells, but the output contact 210 is made sufficiently large
that a quasi-continuum of states is available. The quantum well 202 is
coupled from beneath to an electrode 206, and the quantum well 204 is
coupled from beneath an electrode 208. These are preferably degenerately
doped semiconductor regions, or may be metal lines, but in any case
provide the long-distance routing necessary for formation of conventional
electronic circuits.
In the presently preferred version of this embodiment, the quantum wells
202 and 204 have only 2 components of momentum discretely quantized, since
they are directly connected to their respective electrodes. That is, the
electrodes 206 and 208 can be, for example, tungsten, with a thin layer of
a conventional barrier metallization optionally provided on top. The GaAs
of quantum wells 202 and 204 are formed directly on top of this conductor.
Preferably the electrodes 206 and 208 are n+ GaAs. Optionally, a thin
barrier of AlGaAs can be provided beneath each of the quantum wells 202 or
204 and above the respective contact 206 or 208. This barrier is
sufficiently thin that it is readily tunnelled through, and therefore does
not prevent DC coupling of each quantum well to its respective electrode,
but even this thin a barrier would be sufficiently small to provide
sufficient quantization of all 3 parameters of momentum within each of the
wells 202 and 204, and thereby increase the relative resonant tunneling
gain at the expense of a small increase in processing complexity and small
decrease in overall current.
A larger scale version of essentially this structure is shown in FIG. 9.
This 3-terminal quantum well device can be configured with multiple chains
of quantum well pairs 202, 204 in parallel, for greater current flow. Note
that electrode 206 can be thought of as acting as a source, electrode 208
can be thought of as a gate electrode, and electrode 210 can be thought of
as a drain electrode. Pattern 213 can be used to define the well locations
202 and 204. FIG. 10 shows a cross section of this structure, including
ground plane 211.
A further variation of this structure is shown in FIG. 11. Note that
additional electrodes 208 prime are used to provide longer chains of wells
202, 204 prime, 204 prime, etc., wherein resonant tunneling will occur
only if the voltages on all f the electrodes 206 and 208 prime jointly
satisfy one single condition (or some one of a small set of conditions).
That is, FIG. 12 shows a sample set of electrode bias conditions wherein
resonant tunneling will not occur, and FIG. 13 shows a different set of
bias conditions wherein resonant tunneling will occur.
This multiple-gate device is particularly useful for a read only memory.
In a read only memory (ROM) embodiment, information is hard-programmed
simply by changing the degree of electrical coupling of a column line to
underlying quantum wells. A sample of this embodiment is shown in FIG. 14.
A particular advantage of this embodiment is that the patterning of the
metal lines need not be exactly aligned to the pattern of the underlying
quantum wells. That is, if the pitch of the metal lines can be made twice
or more that of the quantum wells, and the ROM will still function.
In the sample embodiment shown, column lines 302, 304, 306, etc. are
overlying metal lines. Column line 302 is electrically coupled to row 310
of quantum wells, but not coupled (or less well coupled) to row 312 of
quantum wells. This differential coupling can be accomplished by holes cut
in a field plate, or by a dielectric patterned in multiple thicknesses. A
sample mode of operation of this structure is as follows: A background
potential is defined for every column (i.e., all columns fixed at ground)
such that resonant tunneling occurs through each row of quantum devices
310, 312, etc. When it is desired to read out a column of cells, the
column line 302 for that column is changed to some different voltage. This
different voltage will perturb resonant tunneling in any row to which that
column line is electrically coupled, and thereby disrupt resonant
tunneling in that row. In rows to which this column is not electrically
coupled resonant tunneling will not be disrupted. Thus, by monitoring the
current in a row, the information hard-programmed into the intersection of
the addressed column and the read-out row will be detected.
A sample process for fabricating quantum well devices according to the
present invention will now be described. In particular, fabrication of 3
terminal devices, as discussed above, will be used as an example.
The beginning material is a semi-insulating GaAs substrate with 2 epitaxial
layers. The substrate is preferably chromium doped, e.g. to 10.sup.15 per
cubic centimeter, although this is not necessary. The first epitaxial
layer is n+ GaAs. This layer will provide connections, and is therefore
fairly thick and of a fairly high conductivity, e.g. 5000 angstroms thick
and doped to 10.sup.18 per cubic centimeter or more n type. On top of this
is deposited the thin lightly doped n-type layer which will patterned to
form the actual quantum wells. This layer is doped, e.g., about 10.sup.16
per cubic centimeter n-type, and, in a sample embodiment, is 150 angstroms
thick. (The light doping is used merely so that some carriers are
available. Heavier doping would increase inelastic tunnelling.) FIG. 21
shows an initial structure, having epitaxial layers 404 and 406 atop
substrate 402. Optionally, as discussed above, an extremely thin layer of
AlGaAs could be interposed between layers 404 and 406. This structure is
preferably fab | | |
