EP0623244A1 - Quantum well p-channel field effect transistor, and integrated circuit having complementary transistors - Google Patents

Abstract

This transistor comprises an AlxGa1-xAs (or Alx'In1-x'As) layer and a GayIn1-yAs layer defining, at the level of this latter layer, a quantum well comprising HH-type sub-bands. According to the invention, the thickness of the GayIn1-yAs layer is chosen so that, when a negative voltage (VG) is applied to the gate, it appears in the quantum well of the sub-bands HH1, HH2, HH3, ... separated by an energy such that the sub-bands corresponding to the highest effective masses m*h// are populated by holes in concentration notably lower than that of the holes during the HH1 sub-band, so as to create a regime accumulation of holes in the quantum well and correlatively increase the transconductance of the component. This corresponds, for 25 to 35% indium, to a thickness of GayIn1-yAs comprised between approximately 4 and 6 nm or, for 25 to 30% indium, to a thickness comprised between 6 and 9 nm. It is also possible to provide, to further improve performance, a ternary structure such as AlxGa1-xAs/GayIn1-yAs/AlzGa1-zAs, AlxGa1-xAs/GAAswSb1-w/AlzGa1-zAs or AlxGa1-xAs/GayIn1-yAswSb1-w/ AlzGa1-zAs.

Description

 P-channel field effect transistor with quantum well, and integrated circuit with complementary transistors

The invention relates to components of the p-channel field effect transistor type with quantum wells used either in isolation or associated with n-channel field effect transistors in integrated circuits with complementary transistors.

In this regard, digital CMOS silicon integrated circuits constitute an extremely interesting technology thanks to their low consumption, which can reach values as low as 0.3 μW per bit of memory and which therefore allows large-scale integration of complex digital circuits. .

However, this technology does not make it possible to obtain very fast circuits, due to the relatively long propagation time of the silicon CMOS; the access time of the memories produced in this way is thus of the order of 10 ns.

If greater speed is desired, it is necessary to realize the circuits according to another technology, for example from ECL bipolar transistors on silicon, which allow access times of the order of 1 ns.

However, these circuits consume a much greater power, of the order of hundreds of microwatts per bit.

Another possible way consists in using, instead of silicon, III-V alloys, in particular by making the circuits from MESFETs (MEtal-Semiconductor Field-Effect Transistors: metal-semiconductor field effect transis¬ tors) on GaAs , whose parameters of speed and consumption are at values of the order of 0.8 ns for 500 μW. However, these intermediate values, which only achieve a compromise between speed and consumption, do not allow this technology to find significant industrial outlets.

The fastest circuits that we have been able to make up to now include an AlGaAs / GaAs heterojunction and are made based on HEMTs (High Electron Mobility Transistors), components also known as acronym TEGFETs (Two-Dimensional Electron Gas Field-Effect Tran¬ sistors: two-dimensional electron gas field effect transistors) whose propagation time is around 25 to 30 ps per logic gate. But their high consumption, of the order of 4 mW per door, makes it very difficult to integrate on a large scale-

More recently still, it has been proposed to produce integrated circuits with AlGaAs / GalnAs heterojunction, a particularly advantageous technology because it uses, as in the case of CMOS, complementary transistors, with n and p channel. We can refer for example to D.E. Grider et al., Development of Static Random

Access Memories Using Complementary Heterostructure Insulated Gâte Field Effect Transistor Technology, GaAs IC Symposium Di- gest 1990, p. 143. This technique makes it possible to produce integrated circuits with very low consumption, of the order of 64 μW per gate, with a relatively short propagation time, of the order of

200 ps. We were thus able to produce SRAM 4 Kbits memories with remarkable performances of 3.5 ns of access time and a very low consumption of 183 mW (see DE Grider et al., Delta-Doped Complementary Heterostructure FETs with High Y - Value Pseudomorphic In Ga ₁ , _y As Channels for Ultra-Low Power

Digital IC Applications, IEDM Digest 1991, p. 235).

The object of the invention is to overcome a certain number of imperfections and limitations (which will be explained in more detail below) of the components known to be AlGaAs / GalnAs heterojunction proposed up to now.

More precisely, one of the aims of the present invention is to propose an appropriate choice of thicknesses and epitaxial layers which makes it possible to produce a p-channel transistor of very high transconductance, typically with a transconductance value of the same order than that of an n-channel transistor, thus making it possible to produce complementary circuits whose p-channel transistors have performances equivalent to those of n-channel transistors.

Such was not the case until now, the p-channel transistors having, in the best of cases, transconductances more than four times lower than those of n-channel transistors; we will explain in detail below the physical phenomena which are at the origin of these differences.

According to a first aspect of the invention, a component is proposed incorporating a conventional heterojunction structure.

More precisely, this known structure comprises a hetero-junction formed between a layer comprising a III-V semicon¬ conducting material with a large forbidden band (which is advantageously ^ Ga ^^ As on GaAs substrate, or Al _χ .In _{1. Χ} s on InP substrate) and a layer comprising a III-V semiconductor material with a small forbidden band (advantageously Ga 1 ^ As), this heterojunction defining in the valence band diagram of the heterostructure, at the level of the layer comprising the material with a small prohibited band, a quantum well comprising HH-type sub-bands.

Characteristically from the aforementioned first aspect of the invention, the thickness of the layer comprising the material with a small prohibited band is essentially chosen so that, when a negative voltage is applied to the grid, it appears in the well quantum of the sub-bands HH- _{j ^} and HH ₂ , as well as possibly sub-bands of higher order HH ₃ , ..., these various sub-bands being separated by an energy such as the corresponding sub-bands the highest effective masses m ^ are populated with holes in a concentration significantly lower than that of the holes can plant the sub-band HH ^ so as to create a regime of accumulation of holes with low effective mass m _/ in the quantum well and correspondingly increase the transconductance of the component.

According to a second aspect of the invention, there is provided a component having a modified structure compared to a conventional hetero-junction.

Characteristically of this second aspect of the invention, the component successively comprises, epitaxially grown on a substrate: a first layer comprising a III-V semiconductor material with a large forbidden band (advantageously Al ₂ .Ga. ₂ As on a substrate GaAs); a layer comprising a semiconductor material III-V with small prohibited band (advantageously Ga .nL._As, GaAs- ^ Sb _{] ^} .- ^ or Ga _y In _1. ^. s _w Sb _{1. w} ); and a second layer comprising a III-V semiconductor material with a large forbidden band (advantageously Al _χ Ga _{1. χ} As), the crystalline mesh mismatch between these layers being such that the layer comprising the material with small forbidden band is under uniaxial compressive stress in the plane of the layer and define in the valence band diagram of the structure, at this layer, a quantum well comprising HH type sub-bands. In addition, the thickness of the layer comprising the material with a small forbidden band and the composition of the materials with a large forbidden band are essentially chosen so that, when a negative voltage is applied to the grid, it appears in the well. quantum of the sub-bands HH-L and HH ₂ , as well as possibly sub-bands of higher order HH ₃ , ..., these various sub-bands being separated by an energy such that the sub-bands corresponding to the masses the highest effective m ^ are populated with holes in a concentration considerably lower than that of the holes populating the HH-L sub-band, so as to create a regime of accumulation of holes with low effective mass m ^ in the quantum well and correspondingly increase the transconductance of the component.

For one or other aspect of the invention, when the material with a small band gap is Ga _y ln ^. _y As, the thickness of the layer comprising this material can be essentially between about 4 nm and about 6 nm for a molar fraction of indium (ly) _jn between 0.15 and 0.35, so that the concentration of the holes in the HH ₂ sub-band is significantly lower than that in the HH ^ sub-band or between approximately 6 nm and approximately 9 nm for a molar fraction of indium dy) ι _n of between 0.15 and 0.30, so that the concentration of the holes in the sub-band

HH ₃ is notably lower than that in the sub-bands ΗS .- ^ and HH ₂ .

The invention also relates to an integrated circuit with complementary components of the p-channel and n-channel field effect transistor type transistors, characterized in that it comprises at least one effect transistor p-channel field produced in accordance with the above teachings, so as to lower the average effective mass holes for the p-channel transistor and correspondingly increase the transconductance and the current density thereof.

0

We will now describe in detail the invention, with reference to the appended figures; it will be noted that, in all the figures, the same references always designate identical elements.

FIG. 1 shows the general structure of two transistors, respectively with n-channel and p-channel, the invention applies to the structure of the p-channel component.

FIGS. 2a, 2b and 2c show the shape of the conduction band and the valence band for the structure of FIG. 1, respectively at rest, in the case of the application of a voltage of positive gate (for the n-channel transistor) and in the case of the application of a negative gate voltage (for the p-channel transistor). FIG. 3 represents, with the energy as a function of the wave vector, the valence band diagram of GaAs and GalnAs, in the case of an unconstrained material.

FIG. 4 schematically illustrates the structure of a constrained heterojunction and the nature of the stresses undergone. FIG. 5 is equivalent to FIG. 3, for the constrained material of FIG. 4.

FIGS. 6a and 6b illustrate the shape of the valence band, respectively at rest and with the application of a negative grid tension, of a constrained structure with low aluminum content.

Figures 7a and 7b are homologous to Figures 6a and 6b, for a structure with a high aluminum content.

Figures 8a and 8b show the configuration of the quanti¬ well formed at the level of the GalnAs layer, for two different thicknesses of this layer.

Figures 9a and 9b show more precisely, for two negative voltages of different grids, the position of the different sub-bands present in the quantum well presented in FIG. 8b, in particular by taking into account the triangular deformation of the quantum well when a polarization is applied.

Figures 10a, 10b and 10c show, respectively at rest and for two different negative grid voltages, the shape of the valence bank, as well as that of the quantum well and the sub-bands which appear in this well, in the case of the structures of the prior art.

Figures 11a, 11b and 11e are homologous to Figures 10a, 10b and 10c, for the structure according to the first aspect of the invention.

Figure 12 is homologous with Figure 1, for the structure according to the second aspect of the invention.

Figures 13a, 13b, 14, 15 and 16 illustrate the shape of the quantum well formed at the level of the GalnAs layer in various configurations (more specifically, Figures 13a, 13b and 15 refer to the first aspect of the invention, while Figures 14 and 16 refer, by comparison, to the second aspect of the invention).

In FIG. 1, the starting structure of the invention is shown schematically, which successively comprises:

- a GaAs substrate 1,

A buffer layer 2, also of GaAs, or consisting of a GaAs / AIGaAs stack, but with perfectly controlled chemical (purity) and crystallographic characteristics, over a thickness of 500 nm (this thickness value, like all those which follow, being unless otherwise indicated a typical value given only for information),

- a volume or planar doping (“δ-doping”) 3 with silicon,

A layer 4 of GaAs, approximately 3 nm thick,

- a layer 5 of Ga ^ En ^^ s, 15 nm thick, with a gallium y _{Ga content} of about 0.75 to 0.85 approximately (here as in the following, all the contents are indicated in frac- molar positions),

A layer 6 of A ^ Ga ^^ As, 25 nm thick, with an aluminum content x ^ of about 0.75, and

- a protective layer 7 of GaAs, 3 nm thick. It will be noted incidentally that other III-V alloys than those indicated can be used, for example a layer 6 of Al <In ₁ _ _χ -As on a layer 5 of GalnAs, etc., the contents of the various constituents being chosen so as in particular to adapt the mesh parameters of the various alloys, on an InP substrate. Furthermore, all the layers forming this starting structure are, with the exception of planar doping, undoped layers.

On this structure, n-channel and p-channel field effect transistors can be formed by implanting doped zones, respectively n ⁺ and p ⁺ referenced 8 and 9, penetrating to layer 3, and forming in surface of the source, drain and gate electrodes S, D and G, in itself entirely conventional.

It is thus possible to produce logic circuits with complementary transistors with n and p channel, although, as indicated above, the present invention can also be applied to the production of p channel field effect transistors. considered in isolation, ie in the form of discrete components.

In the case of the n-channel transistor, if a strongly positive voltage is applied to its gate, greater than a threshold voltage V _Tn , it will accumulate electrons in the layer 5 of Ga In _x As, thus forming then an n-type channel. Conversely, in the case of the p-channel transistor, if a strongly negative gate voltage is applied, lower than a threshold voltage V _τ , holes will accumulate in the layer 5 of Ga _y ln ^ _y As , thus forming a p-type channel. FIGS. 2a, 2b and 2c show the corresponding configuration of the conduction band E _c and of the valence band E _v respectively at rest, for a positive gate voltage (therefore corresponding to the n-channel transistor) and for a negative gate voltage (therefore corresponding to the p-channel transistor). V _Q. represents the grid voltage applied (zero, positive or negative depending on the case) and E _F represents the Fermi level. The reference 10 indicates the place where the electrons accumulate in the case of the n-channel transistor, and the reference 11 indicates the place where the holes accumulate in the case of the p-channel transistor. We will show that the respective conduction conditions of these holes and these electrons are significantly different, resulting in the structures produced so far, a very large discrepancy between the properties of n-channel transistors and those of transistors to p channel integrated in the same circuit. Indeed, we know that in GaAs or GalnAs the electrons have a low effective mass m _e , while the holes have a high effective mass m _h (we recall that the effective mass corresponds to a statistical average); in other words, the mobility of the electrons is high, but that of the holes is very low. To remedy this drawback preventing the realization of fast complementary logic circuits, it has been proposed by GC Osbourn et al., Electron and Hole Effective Masses for Two-Dimensional Transport in Strained-Layer Superlattices, Superlattices and Microstructures, Vol . 1, No. 3, p. 223 (1985), to associate GalnAs with GaAs or AlGaAs in order to create a constrained layer of GalnAs having the effect of reducing the effective mass of the holes by implementing complex physical phenomena, which will be briefly explained below .

In an unconstrained material, the valence band of GaAs or GalnAs is split into two bands, the curvature of which is very clearly separated. FIG. 3 represents in the plane {S, K} (energy as a function of the wave vector) the diagram of the valence band in this case: one of the bands, referenced HH, is called “band of heavy holes” , and the other, reference LH, is called “strip of light holes”. We know that the effective mass of the holes is inversely proportional to the curvature of the strip, according to the relationship:

*

= ft ² / (d ² ξ dK?) (1)

where h is the Planck constant, ε is the energy and K is the vector wave.

If we now consider a constrained material, for example, as illustrated in FIG. 4, a thin layer of GalnAs lying between two layers of GaAs or AlGaAs, in the plane parallel to the interface the layer of GalnAs undergoes compression, diagrammatically represented by the arrows 12, while in the perpendicular plane it undergoes a tension, shown diagrammatically by the arrow 13.

On the diagram of the corresponding valence band, illustrated in FIG. 5, these deformations have the effect of separating the valence bands and strongly deforming them in the plane parallel to the interface.

In FIG. 5, the right half-plane corresponds to the wave vector Ky parallel to the interface and the left half-plane to the wave vector K _j _ perpendicular to the interface. Parallel to the interface (that is to say for the right half-plane of the diagram in FIG. 5), the bands are strongly deformed, the “heavy” holes HH becoming light and vice versa for the “ light ”LH which become heavy, while perpendicularly (that is to say for the left half-plane of FIG. 5), the“ heavy ”holes HH remain heavy and the“ light ”holes LH remain light . In other words, the heavy / light character of the holes is inverted in one of the valence bands and not in the other.

The deformation of these valence bands also changes the statistical distribution of hole populations; thus, on average, in a GaAs / GalnAs / GaAs or AlGaAs / GalnAs / AlGaAs system, the holes have a lower effective mass than in an unconstrained GaAs / AlGaAs system. Thus, in such a constrained system, the p-channel transistors exhibit increased mobility and, consequently, better transconductance g _mp . The above-mentioned works by Grider thus mention transconductances g _mp reaching 70 mS / mm

(millisiemens per millimeter) for transistors of 1 μm of gate length. This value is however much lower than their counterparts g _min of transconductances of n-channel transistors, which are of the order of 300 mS / mm, ie a ratio of 4.3 which remains to be overcome if the we want the p-channel transistor to work as well as the channel n.

Very recently, it was proposed by B. Laikhtman et al., Strained Quantum Well Valence-Band Structure and Optimal Parameters for AlGaAs-InGaAs -AlGaAs p-Channel Field-Effect Transistors, J. Appl. Phys., Vol. 70, No. 3, p. 1531 (1991), to further reduce the effective mass of the holes by using quantum wells at the level of a layer of GalnAs lying between two layers of AlGaAs, the compositions of these layers as well as the width of the quantum well (this is ie the thickness of the layer of GalnAs) to be defined so that the level HH ₂ is just at the edge of the quantum well. FIG. 9 of this article shows that very low effective masses, of the order of those of the electrons, that is to say close to 0.07 m ₀ (m ₀ being the mass of the electron) can be obtained, provided that the aluminum content is greater than 0.2; it can be deduced from FIG. 8 of this same article that the authors recommend the use of layers of GalnAs of thickness between 2 and 5 nm, the thickest value corresponding to the lowest aluminum concentration x ^ . More precisely, to x ^ = 0.2 corresponds a thickness of Ga ^ -In ^ _y As of between 4 and 5 n approximately, and to x ^ _j = 0.8, a thickness of the order of 2.5 nm about.

However, these structures all have major drawbacks, both in the case of a low aluminum content and a high content. In fact, in the case of a low aluminum content, the structures have a low discontinuity in the valence band, in other words a low potential barrier ΔE _V. FIG. 6 schematically represents the valence band of such a structure where ^ = 0.2 approximately, at equilibrium (FIG. 6a) and under negative grid polarization (FIG. 6b): in the latter case, ΔE _V is lower at 0.1 V and, with such a low height, the holes pass through this barrier by tunnel effect (arrow 14 in FIG. 6b) and / or by thermionic effect (arrow 15). The transistor cannot therefore function correctly, the tunnel effect and the thermionic effect creating excessively large gate leakage currents. On the other hand, structures with a high aluminum content (x _A1 > 0.7 approximately) have reasonably high potential barrier heights ΔE _V , as illustrated in FIGS. 7a and 7b, that is to say greater than 0.4 V approximately and therefore sufficient for the leakage current by tunnel effect to be negligible. But to arrive at this result, the optimal thickness of the Ga In- _y As layer must, according to the aforementioned work by Laikhtman, fall to a value as low as 2.5 nm for an indium (ly) _{jn content} ≥ 0.25 approximately.

Such a small thickness leads to several difficulties, in particular:

- in this small space, the holes (or the electrons) undergo severe interactions with the two interfaces which confine them, thus significantly reducing their mobility, and,

- on the other hand, the practical crystal growth conditions of the AlGaAs and GalnAs layers being very different, it is extremely difficult to obtain interfaces of perfect quality, and we are then in the presence of a certain roughness interface which further deteriorates mobility, especially if the quantum well is narrow. It can therefore be seen that the teachings of the prior art, when they aim to increase the mobility of the holes and therefore to increase the trans¬ conductance of the p-channel transistors, lead in practice to a dead end, a low aluminum content creating excessively large leakage currents and a high aluminum content leading to a layer that is far too thin to be effective and conveniently achievable.

The object of the invention is to overcome all of these limitations, by nevertheless making it possible to produce p-channel transistors of very high transconductance, typically of transconductance as high as that of n-channel transistors.

It can be carried out in two ways which will now be described in detail, one by retaining a conventional structure of ^• quantum well, the other by modifying this structure to further improve its performance. First form of implementation of the invention

The invention incorporates the basic structure which is that illustrated in FIG. 1, with a quantum well formed between a layer 6 of a layer 5 of Ga Jii. _y As and a layer 4 of GaAs. The physical effects occurring in these layers are:

- the effect of mechanical stress on the position and shape of the valence band and on the effective masses of the holes in the plane of layer 5 of GaJ ^. ^ S and in the plane parallel to it,

- the appearance of quantum sub-bands (levels) in the thin layer 5, and

- the tunnel effect through the potential barrier constituted by layer 6 of ^ Ga ^^ s. s.

To these three effects are added the restrictive considerations on the width of the quantum well, that is to say on the thickness of the layer 5 of Ga -n- _y As as was recalled above.

Let us consider this last parameter (quantum well width), taking into account only compositions for which the aluminum content x ^ _j is greater than about 0.7 due to the tunnel effect undergone by the electrons in the transistor at channel n, as has been demonstrated (in this regard, reference may be made to French patent application 91-15140, in the name of the Applicant, which describes and explains this phenomenon in detail).

Let us first consider, as a first approximation, that the quantum well created by Ga _y ln ^ _y As is symmetrical, as illustrated in Figures 8a and 8b. H appears inside this well of the quantum sub-banks HH _1} HH ₂ , HH ₃ , etc. which are, as can be seen in FIG. 8, all the more separated in energy the smaller the width of the well. We will negate the contribution of the LH holes, because the energy levels of the LH _χ and LH sub-bands are very far away (more than 150 mV) from the top of the valence band as soon as the indium (ly) ι _{n content} exceeds 0, 20, which is the case here; these sub-bands are therefore not populated with holes under normal working conditions vail of the transistor.

We know that the effective mass my _/ , in the plane parallel to the interface, of the holes HH ₂ , HH ₃ , etc. is greater than that of HHL holes. It is therefore advantageous for the sub-bands HH ₂ , HH ₃ , etc. are the least populated possible in holes. The aforementioned article from

Laikhtman proposes to use a quantum well sufficiently narrow so that the HH ₂ sub-band is located just at the edge of the well as illustrated in FIG. 8b, which leads to wells of the order of 2.5 nm in width. . But this condition is neither necessary nor sufficient.

Furthermore, it has already been indicated above that this condition was not sufficient when the aluminum content ^ _j is low, due to the tunneling effect of the holes through the potential barrier ΔE _V. This implies to provide contents x _A] > 0.7 approximately due to the risk of tunnel effect undergone by the electrons in the n-channel transistor. However, under these conditions, ΔE _V is greater than 500 mV and, so that the level HH ₂ is not filled with holes, it is not necessary that this level is at the edge of the well, that is to say 500 mV from the bottom of the well; it is only necessary that it is sufficiently distant from the sub-band H ^ so that the ratio of concentration of holes between HH ₂ and HH- ^ is negligible.

If we assume that the negligible limit is a few percent for this ratio, we can then determine the difference in energy level between H ^ and HH so that the concentration of holes in the sub-band HH ₂ is a few percent of that in the subband HH- ^

We know that the concentration p of the holes varies with the energy level E _H of the sub-band according to the relationship:

p = N _v exp [- (E _H - E _F ) / kT], (2)

with:

N _v = m ^* _{h //} kT / π h ² , (3) where E _F is the Fermi level, k is the Bolztmann constant, T is the absolute temperature, h is the Planck constant and my _f is the effective mass of the holes.

The calculations then show that the aforementioned condition (that the concentration of HH ₂ holes is only a few percent of that of HH-L) is fulfilled when the HH ₂ level is about 125 mV of HH-L, which gives quantum wells of approximately 5 nm wide - value much higher than the value of 2.5 nm recommended by Laikhtman. More precisely, in the hole accumulation regime (that is to say with a strong negative polarization of the grid, for V _Q. <-IV approximately) the quantum well is not of rectangular shape, but of pseudo-triangular shape, as illustrated in FIGS. 9a and 9b.

Under these conditions, the energy which separates the sub-levels HH-L ^e ^ HH-2 is a little higher than for the rectangular configuration, but this variation is negligible compared to the differential value of 125 mV.

The pseudo-triangular form is however of great importance in the case where the quantum well is not symmetrical, which is not considered by Laikhtman, but by Grider in the aforementioned article, and whose band diagram has been explained with reference to Figures 2a to 2c.

FIGS. 10a, 10b and 10c show in more detail the diagram of the valence band in this case, respectively for a zero gate voltage and two increasing negative gate voltages. Note that the discontinuity of the valence band ΔE _V between Ga -n ^^ As and GaAs is of the order of 80 to 130 mV, depending on the concentration of indium (ly) ι _n , which can vary between 0.25 and 0.40. We note on the other hand that, at equilibrium (Figure 10a), three quantum sub-bands HH-L, HH ₂ and HH ₃ are present in the well given the relatively large width of the latter (15 nm ). At equilibrium, these wells are filled with holes but lors- ^'as the gate voltage is sufficiently high (Figure 10b), the strip of Bure cour¬ shows a pseudo-triangular wells with pre sence many HH-L, HH ₂ , HH ₃ , HH ₄ quantum sub-bands and HH ₅ . Given the position in energy of these subbands with respect to the Fermi level and HH-L level, ^ ^a subband HH and _2, partially, the subband HH ₃ are filled with holes. However, as described by M. Jaffe et al., Theoretical Formalism to Understand the Rôle of Strain in the Tailoring of Hole Masses in p-Type In _y Ga ^. _y As (on GaAs Substrates) and

I ⁿ 0.53 ₊ χG ^a o.47-o ^s (° ⁿ I ⁿ P Substrates) Modulation-Doped Field- Effect Transistors, Appl. Phys. Lett., Vol. 51, No. 23, p. 1943 (1987) or by Laikhtman in the aforementioned article, the effective mass of the holes of the upper sub-bands HH ₂ and HH ₃ is clearly greater than that in the sub-band HH _j , the mass ratio being able to go up to a factor of ten.

If we take into account the filling statistic of the HH ₁ levels _? HH ₂ , HH ₃ , ..., we can deduce, according to Jaffe, an average effective mass m ^' _/ holes of the order of 0.324 m ₀ , value 4.8 times greater than the effective mass of the electrons.

This result is consistent with the experimental results observed on the transconductance of p-channel transistors, which is 70 mS / mm, value 4.3 times lower than that of n-channel transistors which is 300 mS / mm as indicated above.

As demonstrated above, this drawback disappears if quantum wells having a width of approximately 5 nm are used.

Figures 11a to 11e, which are homologous to Figures 10a to 10c but for such a quantum well width according to the invention, show the shape of the corresponding valence band.

We can see that, at equilibrium (FIG. 11a), only the HH-L sub-band is present, due to the double combined effect of the narrowness of the quantum well and of the low value of ΔE _V between GaAs and GalnAs, of the order of 100 mV. For a sufficient polarization of the grid, of the order of −1 V, (FIG. 11b), the band curvature reveals the sub-band HH ₂ ; due to the narrowness of the quantum well, the latter is then located about a hundred millivolts from HH-L (conversely, if the quantum well was wide, for example 15 nm as in the case described by Grider, the under- HH ₂ band would then be only 50 mV from HH-L and would then fill with holes of large effective mass). For higher polarization values, for example V _Q. = -1.4 V approximately (figure 11b), an HH ₃ sub-band may appear, but its level, like that of HH ₂ , will be sufficiently far from the Fermi level

Ep so that it is not populated with holes if the quantum well is sufficiently narrow.

Thus, even when the quantum well is asymmetrical, the ré¬ reduction of the width thereof (typically to a width of 5 nm) makes it possible to populate the first subband HH-L, ^ ^es others being very weak populated. In summary, when the quantum well has a width of 5 nm, the p-channel transistor only works with holes of low effective mass.

The value of this effective mass is all the lower when the layer 5 of Ga j- ^ „As is constrained, in other words rich in in¬ dium. We know, however, that this constraint must not exceed a certain limit, beyond which dislocations appear at the interface of the layers. (layers 5 and 4) and (layers 5 and 6). In fact, this phenomenon depends on the thickness of layer 5 of (see for example

JW Matthews and AE Blakeslee, J. Cryst. Growth, Vol. 27, p. 118 (1974)). The thinner this layer, the higher its indium content, and the lower the effective mass of the holes populating the HH- _L sub-band. With a thickness of 5 nm, the indium (ly) ι _{n content} can reach approximately 0.35, the effective mass my _j of the holes of the HH-L level then being approximately 0.07 m ₀ , which is practically the effective mass of electrons.

It will be noted that the difference in energy level between HH-L ^e ^ HH ₂ increases with the concentration of indium and that this property therefore favors the decrease in the density of the holes in HH ₂ compared to HH-L.

With an effective mass my _/ of the order of 0.07 m ₀ , the p-channel transistor has a transconductance similar to that of an n-channel tran¬ sistor. It is then possible to produce integrated circuits with complementary transistors thus paired, with a heterojunction Al _χ Ga ₁ _ ^ s / G --- _y In ₁ _ ^. S, with x _A1 > about 0.7 and (ly) _In > 0.25 envi¬ ron, in which layer 5 of Ga _y In _L _ ^ s has a thickness of about 5 nm, which are faster than those manufactured up to now, where this layer had a thickness of about 15 nm. Thus, with a thickness of 5 nm for layer 5, the p-channel transistor works essentially only with holes HH ^ Likewise, the n-channel transistor only works with electrons of the subband E _j ; the E ₂ sub-band located at 170 mV EL is only very sparsely populated. The concentration of the holes in HH _j is limited by the density of states of this sub-band, and likewise the concentration of the electrons in EL is limited. The current density of the transistors, which is proportional to the concentration of the free carriers (holes for the p-channel transistor, electrons for the n-channel transistor), is therefore limited if only the sub-band HH-L ^is ^ P ^eu Placement of holes and the sub-band E _1? of electrons. There is therefore a certain compromise to be defined between low effective mass my / and high concentration of holes (and electrons). With this in mind, it can be accepted for the p-channel transistor to have the HH ₂ sub-band populated with holes whose effective mass is approximately 2 to 3 times higher than that of the HH ^ sub-band while excluding any stand in the HH ₃ sub-band where the effective mass is significantly higher (7 to 10 times). In this case, the width of the GalnAs quantum well is chosen so that the energy difference between HH ₃ and HH-L exceeds 125 mV, the difference between HH ₂ and HH-L being of the order of 40 mV .

These conditions are satisfied for a quantum well width less than approximately 9 nm; for this width of quantum well, the sub-band E ₂ is at 50 mV from E ^ which allows it to be populated with electrons under strong grid polarization. However, the thickness of the layer being increased to 9 nm, it is necessary to lower the indium content of this layer in order to avoid the appearance of interface dislocations; the content of indium (ly) _jn is therefore limited to values of the order of 0.25 to 0.30.

Furthermore, as indicated above, other III-V alloys than those indicated in the example just described can be used, the contents of the various constituents being chosen so as in particular to adapt the mesh parameters of the different alloys. One can thus for example consider the following structure: substrate 1 InP layers 2 to 4 Al _{x '} In-L_ _χ As adapted in mesh parameter on InP, that is x' ^ = 0.48, layer 5 a-Jn ^ _y As under mechanical stress, either y _Ga of the order of 0.12 to 0.22, layer 6 Al ^ In ^ As adapted as a mesh parameter, layer 7 Ga ïn-L. As adapted as a mesh parameter, y ' _Ga = 0.47.

Second form of implementation of the invention

In Figure 12, there is shown schematically the basic structure used in this second embodiment of the invention. This structure is no longer, as in the case of the first embodiment, a binary, conventional, heterojunction structure (a layer of Ga -L. _Y As with a small prohibited band associated with a layer of Al _χ Ga-L_ _χ As with a large forbidden band, the whole being epitaxial on the substrate GaAs), but a ternary structure comprising a layer 5 of GaJJi-L. _y As with a small forbidden band sandwiched between two layers 6 of Al _χ Ga-L_ _x As and 16 of Al _z Ga _1.z As with a large forbidden band, these three layers being epitaxialized on the substrate 1, 2 of GaAs. The resulting structure is therefore that illustrated in FIG. 12, the numerical references identical to those of FIG. 1 designating similar parts, which will not be explained in more detail.

We will now explain the reasons for this structural modification and the advantages that it provides over the structure of the first embodiment of the invention.

To this end, reference is made to FIGS. 13 to 16, which show, in an enlarged manner, the quantum well formed in the valence band diagram with grid polarization V of the order of -1 V. FIG. 13a, which is in fact an enlarged detail of FIG. 11b, shows that, in the first embodiment of the invention, for a quantum well of the order of 5 nm, at the voltage of polarisa¬ tion indicated the HH ₂ sub-band is only a hundred millivolts from the HH-L ^ ^U '^ sub-band provided that the indium (ly) ι _n concentration reaches approximately 0.35.

But a layer of Ga .nL.yAs with (ly) ι _n = 0.35 approximately is as¬ sez difficult to achieve, metallurgy being delicate due in particular to the difference in mesh parameter which becomes very large.

One could then be tempted to design a structure with a layer of Ga _y In ₁ _ _y As less rich in indium. However, for a Ga In-L As material with a low indium content, and for the same quantum well width, it can be seen in FIG. 13b that, for example for (ly) ι _n ≈ 0.20, the discontinuity of the strip of valence between

GaAs and G a In ^^ As is no longer than 65 mV, the level of the HH ₂ sub-band then approaching that of the HH sub-band. under these conditions, it is no longer possible to neglect the concentra¬ tion of holes in HH ₂ , or even in HH. To overcome this limitation, the invention, in its second form of implementation, proposes to push the HH ₂ and HH ₃ sub-bands towards the highest energies by increasing the discontinuity of the valence band by using the place of GaAs of a material with a large prohibited band such as Al ₂ .Ga. ₂ As. The valence band diagram of such an Al _χ Ga _{1 system} . _x Ace /

G DD-1-L. f A ^ Ga- _{L. z} As is illustrated in FIG. 14: it is easy to see that, even for low concentrations of indium, if the concentration z _A] of aluminum of Al ₂ .Ga. ₂ As is sufficiently high, for example z _Aj = 0.30, the discontinuity of the valence band between Ga-JJ- ^ _.y As and Al _z Ga ₂ _ _z As is sufficient for the concentration of holes in the sub -HH ₂ and HH ₃ bands is negligible.

We therefore see that, in this second form of implementation, the indium content is not at all a critical parameter, and that one can choose for the Ga _j .In ₂ .yAs layer a material with low indium content, which metallurgy will be much easier. In addition to the indium content of the its thickness - which defines the width of the quantum well - can also be chosen from a much wider range than for the first embodiment of the invention. Indeed, if we consider FIG. 15 which represents the dia¬ gram of valence band of the first embodiment for a quantum well width of the order of 6 to 9 nm, still with V = -1 polarization V, the HH ₂ sub-band is then approximately 50 mV from HH ₂ and the HH ₃ sub-band is, as in the previous case of FIG. 13a, in the enlarged part of the quantum well.

However, although this HH sub-band is 40 to 80 mV of HH ₂ (depending on the depth of the quantum well, that is to say according to the content of indium (ly) ^, which must be of the order from 0.25 to 0.30), this sub-band is still too close to HH ₂ so that its concentration in holes can be neglected.

To overcome this limitation, in the same way as in the case of FIG. 14, the use in place of GaAs of a material with large prohibited band such as Al _z Ga ₂ _ _z As, allows, as illustrated in FIG. 16 , to increase the band discontinuity and therefore to repel the HH ₃ sub-band to at least 100 mV of HH ₂ , whatever the indium content of Ga. _n ₂ . _y As.

Here again, the aluminum content z ^ _j of Al _z Ga ₂ _ _z As is not critical, it may for example be 0.30, typically z ^ _j > 0.15. It will be noted that the alloy compositions given above are not limiting, and that, in particular for the central thin layer, alloys other than Ga L ^ _y As can be used advantageously.

This is the case with alloys based on antimony, which exhibit greater mobility of holes than Ga Li _ _y As. In this respect, the compound GaSb having a lattice parameter similar to that of InAs, the GaAs alhages _w Sb ₂ . _w (with w on the order of about 0.85 to 0.90 approximately) and Ga _y In ₂ . _y As _w Sb _1.w (with y of the order of approximately 0.70 to 0.80 and w of the order of approximately 0.85 to 0.90) will also be under uniaxial compression stress.

Claims

1. A component of the p-channel field effect transistor type, with a heterojunction formed between a layer (6) comprising a III-V semiconductor material with a large forbidden band and a layer (5) comprising a III semiconductor material -V with small prohibited band whose crystalline mesh mismatch with the rest of the structure is such that this layer comprising the material with small prohibited band is under uniaxial compressive stress in the plane of the layer, this heterojunction defining in the diagram of valence band of the heterostructure, at the level of the layer comprising the material with small forbidden band, a quantum well comprising HH type sub-bands, component characterized in that the thickness of the layer comprising the material with a small forbidden band is essentially chosen so that, when a negative voltage (V) is applied to the grid, it appears in the quantum well of the -HH2 and HH ₂ bands, as well as possibly higher order sub-bands HH ₃ , ..., these various sub-bands being separated by an energy such that the sub-bands corresponding to the effective masses ni ^' _} , // the highest are populated with holes in a concentration significantly lower than that of the holes populating the HH ₂ sub-band, so as to create a regime of accumulation of holes with low effective mass my / in the quantum well and increase correlatively the transconductance of the component.

2. The component of claim 1, in which the material with a large band gap is Al _χ Ga ₂ _ _χ As and the material with a small band gap is Ga _y In ₂ _ _y As, these materials being epitaxially grown on a substrate (D of GaAs.

3. The component of claim 1, wherein the material with a large band gap is Al _χ In ₂ . _χ As and the material with small forbidden band is Ga In _j . _y As, these materials being epitaxial on a substrate (D of InP.

4. A component of the p-channel field effect transistor type, characterized:

- in that it comprises successively, epitaxially grown on a sub¬ stratum (1): • a first layer (16) comprising a III-V semicon¬ conducting material with a large prohibited band,

A layer (5) comprising a m-V semiconductor material with a small forbidden band, and

• a second layer (6) comprising a III-V semiconductor material with a large prohibited band, the crystalline mesh mismatch between these layers being such that the layer comprising the material with a small band prohibited is under uniaxial compression stress in the plane of the layer and define in the valence band diagram of the structure, at the level of this layer, a quantum well comprising HH-type sub-bands, and

- in that the thickness of the layer comprising the material with a small forbidden band and the composition of the materials with a large forbidden band are essentially chosen so that, when a negative voltage (V _Q. ) Is applied to the grid , it appears in the quantum well of the sub-bands HH ₂ and HH ₂ , as well as possibly sub-bands of higher order HH ₃ , ..., these various sub-bands being separated by an energy such that the sub-bands corresponding to the highest effective masses my / are populated with holes in a concentration considerably lower than that of the holes populating the HHL sub-band, so as to create a regime of accumulation of holes with low effective mass ιoa ^* y / in the quantum well and correspondingly increase the trans- conductance of the component.

5. The component of claim 4, in which the material with a large prohibited band of the second layer is Al _j .Ga ₂ .xAs, the material with a small prohibited band is Ga - ^ _y As and the material with a large band prohibited from the first layer is Al _z Ga ₂ _ ₂ As, these materials 23

being epitaxially grown on a GaAs substrate.

6. The component of one of claims 1 and 4, wherein the material with small band gap is GaAs _w Sb ₂ _ _w .

7. The component of one of claims 1 and 4, wherein the material with small band gap is Ga In2_y s _w Sb2_ _v .

8. The component of one of claims 2 and 5, wherein the thickness of the layer comprising Ga L ^. _y As is essentially between about 4 nm and about 6 nm, for a molar fraction of indium dy) ι _n between 0.15 and 0.35, so that the concentration of the holes in the HH ₂ sub-band is significantly lower than that in the HH2 sub-band.

9. The component of one of claims 2 and 5, in which the thickness of the layer comprising Ga _y In _ _y As is essentially between approximately 6 nm and approximately 9 nm, for a molar fraction of indium (ly ) _jn between 0.15 and 0.30, so that the concentration of the holes in the sub-band HH ₃ is significantly lower than that in the sub-bands HH ₂ and HH ₂ .

10. An integrated circuit with complementary components of the p-channel and n-channel field effect transistor type, characterized in that it comprises at least one p-channel field effect transistor produced according to one of the preceding claims, so as to lower the average effective mass m / of the holes for the p-channel transis¬ tor and correspondingly increase the transconductance and the current density thereof.