GB2331841A

GB2331841A - Field effect transistor

Info

Publication number: GB2331841A
Application number: GB9725189A
Authority: GB
Inventors: Timothy Ashley; Anthony Brian Dean; Charles Thomas Elliott; Timothy Jonathan Phillips
Original assignee: UK Secretary of State for Defence
Current assignee: UK Secretary of State for Defence
Priority date: 1997-11-28
Filing date: 1997-11-28
Publication date: 1999-06-02
Also published as: US20020014633A1; JP5325198B2; GB2346481B; CN1284204A; GB2346481A; JP2001525615A; US6624451B2; JP2011049599A; CA2311778A1; WO1999028975A1; KR20010032538A; GB2331841A9; EP1034568B1; KR100542963B1; GB0011379D0; JP2009194392A; GB9725189D0; CA2311778C; EP1034568A1

Abstract

A field effect transistor (FET) is of the type which employs base biasing to depress the intrinsic contribution to conduction and reduce leakage current. It incorporates four successive layers (102 to 108): a p<SP>+</SP> InSb base layer (102), a p<SP>+</SP> InAlSb barrier layer (104), a n intrinsic layer (106) and an insulating SiO 2 layer (108); p<SP>+</SP> source and drain regions (110, 112) are implanted in the intrinsic layer (106). The FET is an enhancement mode MISFET (100) in which biasing establishes the FET channel in the intrinsic layer (106). The insulating layer (108) has a substantially flat surface supporting a gate contact (116). This avoids or reduces departures from channel straightness caused by intrusion of a gate groove, and enables a high value of current gain cut-off frequency to be obtained. In FETs with layers that are not flat, departures from channel straightness should not be more than 50 nm in extent, preferably less than 5 nm.

Description

1

FIELD EFFECT TRANSISTOR

2331841 This invention relates to a field effect transistor (FET). More particularly, although not exclusively, it relates to FETs such as 1V1ISFETs made from narrow bandgap semiconductor materials, ie bandgap EG in the region of or less than 0. 5 eV. It is also relevant to FETs made fl---orn wider bandgap materials for use at elevated temperatures.

Narrow bandgap semiconductors such as indium antimonide (InSb) have useful properties such as very low electron effective mass, very high electron mobility and high saturation velocity. These are potentially of great interest for ultra high speed applications. InSb in particular is a promising material for fast, very low power dissipation transistors, because its electron mobility p. at low electric fields is nine times higher than that of GaAs and its saturation velocity v.t is more than five times higher, despite GaAs having good properties in these respects. InSb is also predicted to have a large ballistic mean free path of over 0. 5 pm. This suggests that InSb has potential for high speed operation at very low voltages, allowing low power consumption, which would make it ideal for portable and high-density applications. Some of the properties of Silicon, GaAs and InSb at 295 K (ambient temperature) are compared in Table I as follows:

Table 1: Properties of InSb at 295 K Parameter Silicon cram InSb Units 5-, Bandgap 1.12 1.43 0.175 eV m. Electron Effective Mass 0.19 0.072 0.013 mo g. Electron Mobility 1,500 8,500 78,000 cm 2 Y1 SA v.t Saturation Velocity 1X107 IX107 >5x 107 cm S-1 , Electron Mean Free Path 0.04 0.15 0.58 gm ni Intrinsic Carrier Concentration 1.6x1010 1. X 107 1.9X 1016 CM-3 2 Until recently, the potentially valuable properties of InSb have been inaccessible at ambient temperatures due to its low band-gap and consequently high intrinsic carrier concentration (-2xlO'6 CM-3), which is six and orders of magnitude above those of Si and GaAs respectively. This leads to InSb devices exhibiting high leakage currents, as the minority carrier concentration at 295K is much greater than the required value at normal doping levels. It was thought for many years that this was a fundamental problem which debarred InSb and other narrow bandgap materials from use in devices at ambient temperature and above. The problem was however overcome by means of the invention the subject of US Pat. No. 5,382,814, which discloses a nonequilibrium metal-insulator- semiconductor field effect transistor (NHSFET) using the phenomena of carrier exclusion and extraction to reduce carrier concentrations well below equilibrium intrinsic levels. This prior art NflSFET is a reverse- biased p+V'7m+ structure, where p denotes an InSb layer, V is a strained In,-.,Al,,Sb layer (underlined p indicates wider band-gap than p), 7c indicates a weakly doped p-type region that is intrinsic at ambient temperature, and the + superscript indicates a heavy dopant concentration; these four layers define three junctions between respective adjacent layer pairs, ie p-p-, p'n and 7cn+ junctions respectively. The active region of the device is the 7c region, and minority carriers are removed from it at the 7rn' junction acting as an extracting contact. The p'.z junction is an excluding contact inhibiting re-introduction of these carriers. In consequence, under bias applied to the device the minority carrier concentration falls, and the majority carrier concentration falls with it to preserve charge neutrality. This produces carrier concentra:ions below intrinsic levels. A similar effect is produced by cooling. Here the expression "intrinsic" is used in accordance with its normal construction to mean that carriers arise largely from activation of valence states, and approximately equal numbers of minority and majority carriers are present in the semiconductor material. This expression is sometimes wrongly used for extrinsic material (eg Si) to indicate simply that the doping level is low, whereas in extrinsic material carriers arise largely from activation of either donor or acceptor states and one type of carrier (electrons or holes) predominates.

3 The device disclosed in US Pat. No. 5,382,814 was a 1 gm recessed-gate enhancement-mode 1v11SFET structure. For investigation purposes a variety of devices of this kind were produced. It was predicted theoretically that the frequencyfT at which the current gain would fall to unity in a device of this kind would be 55 GHz, but measured values were obtained which were only in the region of 10 GHz. The value offT is treated as a figure of merit by those skilled in the art of high frequency transistors. The best value offT obtained for any of these devices was 17 GHz, despite attempts to limit device capacitance associated with overlap of gate contact metal on to source and drain regions. This indicates that it is difficult to realise the full high frequency potential of InSb TvfISFETs.

It is an object of the invention to provide an alternative form of FET capable of exhibiting an improved value of current gain cutofffrequencyfT.

The present invention provides a field efflect transistor (FET) of the kind comprising biasing means for depressing the intrinsic contribution to the charge carrier concentration in an intrinsic region thereof, characterised in that the FET includes means for defining a channel extending between a source region and a drain region with any intervening departure from channel straightness being not more than 50 mn in extent, as appropriate to enable a high value of current gain cut-off frequency to be obtained. Any such departure from channel straightness is preferably not more than 5 nm in extent.

Here the expression "extent" is to be construed as the maximum height differential between any two regions of the channel, eg its central region and a region adjacent to the source or drain.

The invention provides the advantage that it is capable of providing greatly enhanced values of current gain cut-off frequency compared to the prior art, indicating greatly improved high frequency performance. vflSFETs in particular in accordance with the prior art were found to have disappointing performance at high frequency much below theoretical expectations. The

4 reason for this was originally not understood. However, a number of hypotheses were investigated in an attempt to resolve the problem. One of these hypotheses was that over-etching a NfiSFET gate recess might degrade high frequency performance. Devices of the invention produced without an intervening gate groove intrusion have exhibited much better performance at high frequency, and it is inferred that the hypothesis of the deleterious effect of gate grooving on performance is confirmed.

In one aspect, the FET of the invention is an enhancement mode NUSFET; it may incorporate source and drain regions which are produced by introduction of heavy doping into a layer incorporating the the intrinsic region The source and drain regions may be produced by r) implantation, diffusion doping, alloying or introduction of damage. The intrinsic region may be residually p-type doped and form extracting contact means with the source and drain regions, the channel formed in the intrinsic region in response to bias being n-type.

In a preferred embodiment, the intrinsic region has an interface with a barrier region itself having an interface with a base region, and the intrinsic, barrier and base regions (106, 104, 102) being of like conductivity type and the barrier region being of relatively wider bandgap than the intrinsic and base regions and providing an excluding contact to the intrinsic region.

The FET of the invention may include a gate contact insulated from and extending at least over that pan of the intrinsic region between the source and drain regions to define an enhancement channel therebetween in operation. The base region may be of p' InSb with a dopant concentration r,X 10 17 CM-3 of at le-si - - the barrier region may be of p' In,,Al.Sb with x in the range 0.05 to 0.25 with a dopant concentration of at least 5x 1 017 Cm -3; the intrinsic region may be of 7r InSb with a dopant concentration of less than 5x 1017 CM -3, preferably 1 X 10 15 Cm- 3 to 5x 1016 Cm-3 and the source and drain regions may be of n' InSb with a dopant concentration of at least 5x 1017 Cm-3.

The base, barrier and intrinsic regions are preferably successively disposed in a layer structure, the source and drain regions being produced by implantation, diffusion alloying or damage in the intrinsic region, and the intrinsic region preferably has a substantially flat surface portion supporting a gate insulation layer and a gate contact.

In another aspect, the FET of the invention is a depletion mode NUSFET having an associated channel region. It may incorporate source and drain regions which are heavily doped outgrowths formed upon either the intrinsic region or the channel region; these regions may alternatively be produced by implantation, diffusion, alloying or introduction of damage. They may define therebetween a gate recess accommodating a gate contact.

The intrinsic region may be p-type and either itself or the channel region may form extracting contact means with the source and drain regions.

In a preferred embodiment, the intrinsic region has an interface with a barrier region which itself has an interface with a base region, the intrinsic, barrier and base regions being of like conductivity type and the barrier region being of relatively wider bandgap than the intrinsic and base regions and providing an excluding contact to the intrinsic region. In this embodiment:- 1017 the base region may be of p' InSb with at least 5x acceptors cm-3.

the barrier region may be of 12' In,_.M.Sb with x in the range 0.05 to 0. 25 and at least 5xl 017 CM-3, acceptors the intrinsic region is of n InSb with less than than 5xl 017 acceptors CM-3, preferably in the rangelx 1015 CM-3 to 5x 1016 cm-3; and the source and drain regions are of n' InSb with at least 5x 1 017 donor CM-3.

The intrinsic region may support a channel region, the base, barrier, intrinsic and channel regions being successively disposed in a layer structure, the source and drain regions being grown upon the channel region and the channel region having a substantially flat surface portion supporting a 6 gate insulation layer and a gate contact. The source and drain regions may define therebetween a gate recess, the channel region having a surface portion at an end of the recess supporting the gate insulation layer and gate contact.

The channel region may lie between parts of the intrinsic region, the latter forming extracting contact means in combination with the source and drain regions.

The base, barrier and intrinsic regions are preferably successively disposed in a layer structure, the intrinsic region containing the channel region and supporting the source and drain regions.

Z1) The biasing means for depressing the intrinsic contribution to the carrier concentration in the intrinsic region is preferably arranged to bias the FET at a point of infinite differential impedance where the variation of gate threshold voltage due to substrate bias voltage variations is minimised.

In an alternative aspect, the invention provides a method of making an FET of the kind comprising biasing means for depressing the intrinsic contribution to the charge carrier concentration in an intrinsic -eLIon thereof, characterised in that the method includes defining a channel extending between a source region and a drain region such that any intervening departure from channel straightness is not more than 50 nrn in extent, as appropriate to enable a high value of current gain cut-off frequency to be obtained. Any such departure from channel straightness is preferably not more than 5 nm in extent.

In order -hai the invention might be more fully understood, embodiments thereof will now be described, by way of example only, with reference to the accompanying drawings, in which:- Figure 1 is a schematic sectional view of a prior art NfiSFET not drawn to scale;

Figure 2 shows a gate of the prior art IvfISFET of Figure 1 on an expanded scale;

7 Figure 3 is a schematic sectional view of an n-channel enhancement mode NfiSFET of the invention (not drawn to scale) illustrated in idealised form with layers that are neither concave nor convex; Figure 4 is a schematic sectional view of a central region of an n- channel enhancement mode NfiSFET of the invention showing a minor degree of gate region concavity; Figure 5 is a band structure diagram for a reverse-biased p'p'7rn' NfiSFET structure of the invention; it corresponds to a section on fines V - V in Figure 3; Figure 6 graphically illustrates the output characteristic of the Figure 3 NfiSFET; Figure 7 graphically illustrates the transfer characteristic of the Figure 3 llyfiSFET-, Figure 8 illustrates the variation of current gain cut-ofF frequency with gate length for the Figure 3 NfiSFET and for similar modelled devices; Figure 9 illustrates the variation of AC gain parameters with frequency for the Figure 3 NfiSFET; FigurelO is a schematic sectional view of an n-channel depletion mode NfiSFET of the invention (not drawn to scale); Figurell illustrates the variation of current gain cut-off frequency fT (GHz) as a function of gate length (gm) for a variety of device technologies, both measured and modelled; and Figure 12 is a base currentlvoltage characteristic theoretically achievable for a NfiSFET of the kind in which carder concentration is depressed by exclusion and extraction.

8 Referring now to Figure 1, there is shown a prior art Nff SFET 10 of the kind disclosed in US Pat No. 5,382,814. The NfISFET 10 consists of layers of indium. antimonide (InSb) and indium aluminium antimonide (Inl,AI.Sb). It has a substrate (not shown) supporting four layers of semiconductor material as follows: a heavily doped narrow bandgap p-type (p base layer 12, a relatively wide bandgap heavily doped p-type (p) barrier layer 14, a lightly doped p-type (7C) active layer 16 and a heavily doped narrow bandgap n-type (n) source/drain layer 18. Layers 12, 16 and 18 are InSb and layer 14 is In,-.Al,,Sb. The ir active layer 16 has predominantly intrinsic conductivity at 295K ambient temperature, whereas other layers 12, 14 and 18 have predominantly extrinsic conductivity at this temperature. Interfaces between pairs of adjacent layers 12/14, 14/16 and 16/18 are p'p', 12'7r and 7rn' junctions respectively, the first two of these being heterojunctions and the last a homojunction. The p'7r and 7rn'junctions 14/16 and 16/18 are an excluding contact and an extracting contact respectively.

The N1ISFET 10 has source and drain regions 20 and 22 with contacts 24 and 26 respectively, the regions having an intervening gate recess or groove 28 formed by etching through the n' layer 18.

1 The bottom and sides of the gate depression are covered by a silicon oxide gate insulator 30 and a metal gate contact 32. It is necessary for operation of the N1ISFET 10 that the recess 28 extend complete!,,, (or almost completely) through the n' layer 18 to avoid a short circuit between the source 20 and drain 22.

Carrier eXtraction and exclusion take place in the Nff SFET 10 when bias is applied in normal operation. The lavers 12 to 18 form a p+p'7cn' diode which is reverse biased in operation, ie the I - base layer " 2 is biased negative with respect to the source region 20. Under this bias, the 7rn' junction '1 6/ 18 acts as an extracting contact removing electrons (minority carriers) from then layer 16. Such electrons cannot be replenished from the p+ barrier layer 14, because it acts in combinatiop with the p' base layer 12 as an excluding contact and provides a potential barrier to electron flov,,- to the n layer 16. The electron concentration in the 7c layer 16 therefore falls when the NffSFE- -1- 10 is biased, and with it the hole concentration in that layer for charge neutrality 9 reasons. This greatly reduces the conductivity in the layer 16, which therefore reduces the leakage current between the source 20 and drain 22.

As has been said, trial examples of the N1ISFET 10 have been produced which exhibited disappointing performance at high frequency. The theoretical value of the current gain cut-off frequencyfT (at which the current gain falls to unity) for these devices was 55 GHz, but measured values were only in the region of IOGHz despite attempts to limit device capacitance associated with overlap of gate metal on to source and drain regions. The reason for this was not understood. However, a number of hypotheses were advanced and investigated theoretically in an attempt to identify a candidate artefact responsible for poor high frequency performance. One of these hypotheses was associated with the depth of the gate recess or groove. Theoretical calculations indicated that, if the gate recess was etched too deeply (which is difficult to control and measure), high frequency performance would be degraded. The theoretical situation is illustrated in Figure 2, in which the gate recess 28 is shown on an expanded scale.

For the NIISFET 10 to be viable, the gate recess 28 must extend through most of - preferably entirely through - the n' region 18 to avoid a short circuit between source and drain; if this recess were to be over-etched it would continue down into the 7E region 16, and it is postulated that an inverted step 34 was formed in the trial devices of the order of 100 nm in depth. If so, the NUSFET channel (not shown) between source 20 and drain 22 would be U-shaped instead of straight as intended (in an enhancement mode NIISFET the channel is formed under the gate electrode only in operation in response to application of bias).

To minimise the transit time of charge carriers between source and drain, the source-drain electric field should be directed longitudinally of the channel, which is only possible in that part of a Ushaped channel extending parallel to this field. In such a channel there would be regions in which the channel does not extend in the source-drain direction and the source-drain electric field is therefore inclined to the longitudinal channel direction; in consequence the longitudinal field component would be weaker than would be the case if the field were parallel to this direction, which increases the transit time of charge carriers between source and drain as compared to the case where the field is entirely longitudinal. It is therefore deduced that a U-shaped channel would degrade fT because the latter is related to this transit time. If correct, this theoretical analysis implies that an etched gate recess should be avoided. This hypothesis was tested by the production of examples of the invention hereinafter set out.

Referring now to Figure 3, an enhancement mode MISFET of the invention is indicated generally by 100. It comprises a weakly doped p-type InSb substrate (not shown) upon which are grown in succession first, second, third and fourth layers 102, 104, 106 and 108 having the following thicknesses and compositions:- CM-3.

first (base) layer 102: p' InSb 2 tm thick, Be dopant concentration 3xlO18 2 c 3. second (barrier) layer 104: p' Ino.8sAlo. 15 Sb 20 nm thick, Be dopant concentration 3x 10 18 nf) third (intrinsic) layer 106: Tc InSb 0. 5 Lrn thick, Be dopant concentration lxlO15 Cm7 3; and fourth (insulating) layer 108: Si02 70 rim thick.

More generally, suitable compositions for InSbAnAISb FETs of the invention are as follows:base region: p' InSb with an acceptor concentration of at least 5xl 017 CM "3; barrier region: p' In,,,Al,,Sb with x in the range 0.05 to 0.25 and an acceptor concentration of at least 5xl 017 CM-3; and intrinsic region: n InSb with an acceptor concentration of less than 5xl 0 17 CM-3, preferably in the range IX1015 CM-3 to 5xl 016 CM-3.

The MISFET 100 may optionally include a buried p-type layer 109 (indicated by chain lines) within the channel layer 106 to improve confinement of charge carriers near the insulating layer 108. It has n' source and drain regions 110 and 112 each approximately 0.2 gm thick with a neutral acceptor concentration of at least 5xl 0 17 CM-3. It has source, gate and drain contacts 114, 116 and 118 consisting of successive Cr and Au layers (not shown). The mesa length of the 11 NUSFET 100 is 12 im, this being its full width in the plane of the drawing. It has a mesa (and gate) width of 50 tm, this being the dimension extending perpendicular to the plane of the drawing. The length of the gate contact 116 (horizontal dimension in the plane of the drawing) was nominally 0.7 Lrn; here "nominally" means that lithographic masks were used of appropriate dimensions to produce the required length, but the length was not measured. Other devices of similar construction and type were also produced with nominal gate lengths in the range 0.7 to 2 Lrn and mesa width of 100 tm. The layers 102 to 108 were grown by molecular beam epitaxy. The source and drain regions I 10 and 112 were produced by ion implantation using 70 keV S32 ions with a dose per unit area of 5xl 013 cnf2. The ions were implanted through a native anodic oxide mask at an angle of 10' from the normal to the <110> direction in the crystal with a substrate temperature of 100'C. This was followed by a rapid thermal anneal at 420'C for 10 seconds with an Si3N4 cap to activate the dopant and remove damage. Samples were then anodised and the oxide was stripped off to remove damaged material before applying contacts. The process of producing the NHSFET 100 can result in a minor degree of convexity of the layers 106 and 11-08, but it does not result in a height differential of more than 50 nm between the central region of the channel (when formed in response to gate bias voltage) and outer regions of the channel adjacent the source and drain. The implantation established conducting paths permitting contact i. o the source and drain regions I 10 and 112.

The fourtb layer 108 consisted of 40 nm of sputteredSi02 deposited on top of 30 rum of anodic oxide. Photolytic Si02 would have been preferable for the entire layer 108 but this was unavailable.

The gate -contact 116 extends over the whole of that part of the third or 7C layer 106 which is between he source and drain regions 110 and 112, and overlaps these regions a little. Ideally, to minimise device capacitance the overlap would be zero, but in an enhancement mode device it is importani Sor the gate contact to extend fully between these regions so that in operation the 1 channel region can be established as required.

12 1 In the N1ISFET 100 the direct line between source and drain regions 110 and 112 is not significantly obscured by the geometry of the gate 116, certainly not to the extent of 100 run experienced in the prior art device 10.

Figure 3 is an idealised illustration, in that in practice an FET of the invention may have non-planar layers because of departures from ideal geometry caused by inaccuracies in the production process; ie the intrinsic region may be concave, convex or undulating so long as this does not result in too severe a distortion of the channel. Concavity is illustrated in Figure 4, in which parts equivalent to those described with reference to Figure 3 are like-referenced with a suffix R. Figure 4 shows a central region 120 of an FET equivalent to the N1ISFET 100, except that it approximates morely closely to a practical device. It includes an intrinsic layer 106R, a gate insulation layer 108R and gate contact 11 6R all of which are concave defining a V-shaped recess 122. The bottom of the recess 122 is defined by a vertex 124 at the centre of the insulation layer 108R. The recess 122 has a depth indicated by H which is not more than 50 mn. When the channel (not shown) is established in response to application of gate bias voltage, the recess 122 does not result in a differential of more than 50 rim existing in the vertical direction in the drawing between the central region of the channel (not shown) under the vertex 124 and outer regions of the channel adjacent the source I I OR and drain 1121, in other words any departure from channel straightness arising from gate region non-planarity will be less that 50 wn. Here the gate region is the upper part of the intrinsic layer 106R adjacent the gate insulation layer 108R in which the channel is formed and which determines the channel shape. In other embodiments of the invention it is the active region of the device accommodating the central region of the channel.

The recess depth H and any consequent departure from channel straightness are preferably not more than 5 run. In consequence, the channel when established can extend substantially as 13 determined by the source, gate and drain voltages. As will be described later, this gives rise to greatly improved performance as compared to the prior art device 10.

Figure 5 provides a band structure diagram and associated charge carrier densities in the NfiSFET 100 and as a function of vertical distance x in ptm measured from the lower edge of the first layer 102. The data given in this drawing are for a reverse-biased p'p+7M' structure; they relate to a vertical section through the NfiSFET 100 on fines V - V in Figure 3 extending through the first (p'), second (p') and third (7r) layers 102, 104 and 106 and the (nl source region 110. The right hand ordinate is graduated in IE+12 to M+19, which indicates 1012 to 1019 ern. The drawing shows the following variations, conduction and valence band energies in graphs 140 and 142, net dopant concentration in graph 144, and hole and electron concentrations in graphs 146 and 148 respectively.

The intrinsic earner concentration at 295 K (ambient) in InSb is 2x10'6 CM-3. Graphs 146 and 148 show that in operation the actual carrier concentrations are up to two orders of magnitude less than this throughout most of the activez region of the third layer 106, which corresponds to the approximately flat portion of graph 144. This demonstrates that the carrier concentrations in the 7c region of the layer 106 are being depressed by carrier exclusion and extraction arising respectively from the -3-7-, and 7rn'juncti6ns 104/106 and 106/112.

The NffS'rET 100 was tested in a common-source configuration with the base layer 102 biased to about -0. -: 5 V relative to the source 114 to perform the carrier extraction; this voltage is defined as Vb. and corresponds to the position of maximum dynamic resistance of the diode structure provided ',y the base layer 102 and the source I 10 or drain 112. Drain and gate voltages of normal polarity xere then applied, ie they were both biased positive relative to the source I 10 apart from one case where a small negative gate-source voltage was used.

14 The output characteristic of the 1Y1ISFET 100 is shown in Figure 6. It comprises nine graphs such as 160 and 162 indicating the variation of drain current Idwith drain voltage Vd, at constant gate voltage V,, from 0.2 V to 1,4 V in steps of 0.2 V between adjacent graphs. It can be seen that the drain current starts to saturate at a drain voltage of about 0. 15 V, as indicated by a slight knee in each graph; this is a very low voltage for saturation to initiate, and it is due broadly speaking to the electron mobffity in InSb being very high. It is advantageous because it implies that the N1ISFET 100 will have low power requirements. The output characteristic is generally of the classical form for NfiSFETs, which is evidence that a viable NfiSFET has been produced. At a drain voltage V& of 0. 3 V the drain current is switchable between about 10 and 110 mAmm" by changing the gate voltage V. from -0.2 to 1.2 V.

The transfer characteristic of the N1ISFET 100 is shown in Figure 7. It comprises five graphs such as 160 and 162 indicating the variation of transconductance g,,, with gate voltage V,, at constant gate voltage. The gate voltage changes between adjacent graphs from 0. 1 V to 0.5 V in steps of 0 0. 1 V. It can be seen that the maximum DC transconductance of the Iffl SFET 100 is about 120 mS mm-', and the threshold gate voltage is about 0. 4 V.

The leakage pedestal of the N1ISFET 100 is about 8 rnA mm-', as shown bythe intercept on the current axis in Figure 6. The maximum drain currents are about 120 mA mm-1, determined by the drain current ceasing to increase with gate voltage. The device starts to break down slowly ata drain voltage of around 0.5 V as indicated by the upward curvature of the current/voltage graphs at high drain voltage; this is due to band-to-band tunnelling and surface leakage (probably surface tunnelling generation), both of which it is possible to reduce. The N1ISFET 100 had a significant resistance in series with the channel, of about 2.5 Ohm on each side, measured by forward biasing the base-source/drain diodes. This is believed mainly to be due to the contacting process, and will reduce the transconductances (and hencefT) below ideal values.

is The AC Parameters of a number of enhancement mode NUSFETs of the invention with differing gate lengths were measured by the S-parameter method with a drain voltage of 0. 5 V, and the gate voltage was tuned for maximum S21. Results were de-embedded from parasitic bond-pad capacitances using Koolen's method. Figure 8 shows the measured maximum current gain cut-off frequenciesfT as a function of gate length. The results follow a 1Z dependence, as indicated by a lower line 180, this is as would theoretically be expected if the velocity is not saturated at pinchoff, Values offT as a function of gate length were also calculated (modelled results) for these NUSFETs, and indicated by an upper fine 182; they agree fairly wen with the experimental values, although they do show some effect of velocity saturation, presumably because the channel mobility used was higher. This suggests that further improvement is possible from reduced gate lengths.

The AC parameters of the N1ISFET 100 are shown in Figure 9, The measured current gain cut-off frequency,fT, is 74 GHz - as far as is known at present this is the ffighestfT measured for any FET with 0.7 gm gate length irrespective of transistor type or material. It is more than a factor of four greater than the best value (17 GHz) obtained for any of prior art devices 10 produced for investigation purposes, and more than a factor of seven greater than the typical value (10 GHz) for these devices. It should be possible to increasefT further by reducing the resistance in series with the channel. The frequencyf. at which the unilateral power gain falls to unity is 89 GHz; this is limited by the channel series resistance, and by the output conductance which is relatively low at present. It should therefore be possible to achieve increasedf. in devices of the invention. These values forf. and f. represent a great improvement over the prior art and it is inferred that they provide confirmation of the correctness of the hypothesis of the deleterious effect of gate grooving on perfornance of prior art devices.

16 The results obtained for devices of the invention were very promising, and indicated the potential of the invention to provide high-speed, lowpower devices. Modelling - ie calculation - of device characteristics was performed using ATLAS, a 2D drift-dilTusion device simulator from Silvaco International, using published or measured results for InSb material parameters. The modelled results were tested against experimental p'12'nn' diodes and prior art MISFETs, and found to be in good agreement in terms of leakage currents, transconductance and fT. Modelled results for an implanted enhancement-mode MISFET structures similar to that shown in Figure 3 but with gate lengths of 1 gm and 0.25 pm are shown in the following Table 3, which provides theoretically attainable values for maximum g.,fT andf... The values for 0.25 pm gate length assume the gate insulator oxide thickness to be scaled by the same factor as the gate length.

Table 2: Modelled Parameters for Enhancement-mode vHSFETs of the Invention Gate Length 1.0 pm 0.25 pm Maximum g. 188 mS mm-1 500 mS mm-1 Current Gain Cut-offFrequencyfT 68 GHz 185 GHz Unity Unilateral Power Gain Frequencyf,. 202 GHz 264 GHz These maximum transconductance values represent a considerable improvement on that of 25 mS mm-1 quoted for the prior art device of US Pat. No. 5,382,814.

The MISFET 100 is an InWIni-Al,,Sb heterostructure. There are a number of other semiconductor material combinations that are suitable for construction of devices of the invention. Two semiconductor materials are required with dilfering bandgaps, but they need not be lattice matched. The lesser bandgap should be sufficiently narrow that it is possible to purify the material enough to exhibit predominantly intrinsic conductivity at the FET operating temperature (impossible with Si at the present time); this implies a bandgap which is in the region of or less than 0.5 eV for a device operating at ambient temperature of 295K, but materials of greater bandgap may be used for elevated operating temperatures.

Combinations of materials that may be used to produce FETs of the invention include PbSe/PbS, In,-yAlySb/lnl,AI,,Sb, InAs/InAsi-,,P,,, InAsi,SbjInl_yAlySb, InAsl-,,Sb,,dnAsl-yPy, GaAs/Gal,Al,,As, In,-,Ga,, Sb/Ini-yAlySb and Hgi..Cd.Teffigl_yUyTe. Values of the composition parameters x, or x and y must be suitably chosen. The NfiSFET consisted of InSWIni,Al.Sb, which is a special case of the first of these with the parameter y equal to zero.

Referring now to Figure 10, a depletion mode Nff SFET of the invention is indicated generally by 200. It comprises a weakly doped p-type InSb substrate (not shown) upon which are grown five successive layers 202, 204, 206, 208 and 210 having the following thicknesses and compositions: first (base) layer 202: p' InSb 2 tm thick, Be dopant concentration 3x1O18 rM-3; 018 CM-3 second (barrier) layer 204: p+ Ino.s5Alo.jsSb 20 wn thick, Be dopant concentration 3x I third (intrinsic) layer 206: 7c InSb 0. 5 Lrn thick, Be dopant concentration I x 10'5 Cm-3; and fourth (channel) layer 208: p InSb 20 nm thick, Si dopant concentration 3x1 017CM-3; fifth (- ate insulation) layer 210: Si02 70 nm thick.

The NfiSFET 200 may optionally include a buried p-type layer 211 (indicated by chain lines) within the intrinsic layer 206 to improve confinement of charge carriers near the gate insulation layer 210. It has n' source and drain regions 212 and 214 each approximately 0.2 pim thick with an Si dopani concentration of 2x10'8 cm-3 and forming extacting contacts to the channel layer 208. These reg.ons provide carrier extraction in the intrinsic layer 206 via the channel layer 208. The MSFET 2-00 has source, gate and drain contacts 216, 218 and 220 consisting of successive Cr and Au layers (not shown). Except where indicated above, the NfiSFET 200 has dimensions sin-dlar to the enhancement mode device described earlier. It has a mesa length 12 im, a mesa (and gate) width of 50 ptm and gate contact length nominally 0.7 im. The layers 202 to 208 were is grown by molecular beam epitaxy E). The source and drain regions 212 and 214 were produced by MBE growth on the channel layer 208, the central region of the latter being being masked to avoid growth upon it. After removal of this mask, the gate insulation layer 210 and electrode 218 are deposited. The source and drain regions 212 and 214 define between them a gate recess 222 within which the gate insulation layer 210 and gate contact 218 are located supported by the channel layer 208. The recess may be more or less deep than the height of the gate electrode 218. It is important to note that the gate recess 222 is not a groove produced by etching as in the prior art, but instead a recess defined by growth of upstanding sides. In consequence the formation of the recess 222 does not imply problems associated with too deep a groove affecting channel shape.

The fifth layer 210 consisted of 40 nm of sputtered Si02 deposited on top of 30 nm of anodic oxide. The gate contact 218 extends over most of that part of the channel layer 208 which is between the source and drain regions 212 and 214; the degree to which it does this is not very critical because it is merely required for modulation/depletion of an existing device channel, as opposed to establishment of a complete channel between source and drain which is required in an enhancement mode device.

The NHSFET 200 employs a channel layer 208 to provide an accessible source of electrons; the latter in turn provides a conducting path from source 212 to drain 214 depletable of charge carriers by the gate electrode potential. This path may be entirely within the channel layer 208 or the intrinsic layer 206 or may be partly in one of these and partly in the other. The threshold voltage of the Nff SFET 200 is determined by the doping per unit area of the channel layer 208, ie the product of the layer thickness and its doping per unit volume. The channel layer 208 extends substantially along the direction of the electric field produced by the source-drain voltage in the absence of a gate voltage. This layer is not U-shaped to any unacceptable degree; ie any concavity or convexity in this layer is less than 50 nm in extent.

19 Modelled - ie theoretical - performance figures were obtained for the depletion-mode MISFET 200 assuming that the gate insulation oxide thickness was equal to that in the device 100. These figures are shown in Table 3 as follows:

Table 3: Modelled Parameters for Depletion-mode N1ISFETs of the Invention Gate Length 1.0 gm 0.25 gm Maximum g. 190 ms mm 1 590 mS mm-1 Current Gain Cut-offFrequeneyfT 68 GHz 220 GHz Unity Unilateral Power Gain Frequencyf. 164 GHz 377 GHz Referring to now to Figure 11, current gain cut-olf frequency fT (GHz) is shown as a function of gate length (tm) for a variety of device technologies, actual or modelled, as follows: InSb ideal (calculated from gate length and carrier velocity only), modefled InSb enhancement and depletion mode MISFETs, the InSb enhancement mode WSFET 100, InP- and GaAs-based HEMTs and silicon NMOS.

It can be seen that the Table 2 results forfT in the NfiSFET 100 are only slightly below the ideal trend-line in Figure 11, with a slight tailing of due to the overlap capacitance from the source and drain implants. These results should be anainable with improved performance ftom the implanted diodes, reducCd series resistance and patterning of the gate oxide to reduce overlap capacitance. Comparison of the Table 3 results for fT in the N1ISFET 200 with Figure 11 indicate that this depletion-:-, ode device does not sufFer from the saturation elfect seen in the enhancement-mode devices such as the MISFET 100, and this is attributed to the former having lower input capacitance.

The 1VIISFET 200 may have a different form of channel. The 20 rim thick channel layer 208 may be replaced by a 3 nin thick InSb channel layer with a Si dopant concentration of 2x1018 CM-3 separated from the gate oxide layer 210 by a 7r InSb layer 20 run thick with a dopant concentration Of IX1015 CM-3. This is equivalent to the channel layer being reduced in thickness and buried in the intrinsic 7r layer 206, and is estimated to give a 30% operating speed enhancement. In this case the source and drain regions 212 and 214 form extracting contacts with the intrinsic layer 206.

Referring now to Figure 12, there is shown a base current/voltage I13VBs characteristic 250 which is theoretically achievable by a N1ISFET of the kind in which carrier concentration is depressed by exclusion and extraction. Here the base current is that flowing between a base layer and a source region. This current is not for the purposes of biasing the N1ISFET source, gate and drain relative to one another. Instead it is for the purposes of reducing the carrier concentration and leakage current in the intrinsic device region. The characteristic 250 corresponds to a device containing a lower Shockley/Read trap density than is currently attainable. The device is a p'p'7cn' diode structure which is reverse biased in operation, ie it has a base layer which is biased negative with respect to its source region. Under this bias, carrier extraction and exclusion take place in the intrinsic layer, from which electrons (minority carriers) are removed by an associated 7m'junction acting as an extracting contact.

At a point 252, the slope of the IBVBs characteristic 250 is zero, indicating infinite differential impedance. At this point, the variation of gate threshold voltage with base bias voltage VBs referred to as "back gating" - is minimised, and so this is the preferred operating point for base bias.

21 The modelled results referred to above come from drift-diffusion simulation and neglect ballistic effects, which are expected to become significant at gate lengths of around 0.5 pm. The effect of this will be to increase the average saturation velocity, and hence the g. andfT leading to potential for greater improvements in performance.

Transistors of the invention are potentially applicable to Iiigh-speed analogue device applications. If grown on a semi-insulating substrate or virtual substrate, they could be used in microwave integrated circuits. An InSb device is operative at low voltage, less than 0.5 V, and is therefore characterised by low power consumption which is extremely useful for hand-held applications, providing longer battery lifetimes. Also, it has high electron velocity which allows higher ultimate frequencies to be reached, or alternatively provides the required operating speed at a longer gate length, making it more robust. Transistors of the invention may also be used as digital devices, especially for low-complexity circuits. They are very attractive for high- speed low-power applications because they potentially have a very low Rr product,, where P is energy dissipated in a switching operation and T is the time to switch.

The invention provides FETs which are fast and have low power-dissipation utilising the inherent high electron mobility and saturation velocity of InSWInj,Al.Sb. These FETs give high-speed, low-power performance and demonstrate off-state leakage, currents which are well below levels normally associated with InSb/Ini,Al.,Sb, due to the incorporation of carrier exclusion and extraction techniques. IVfiSFETs of the invention with 0.7 gm gate length have the highest fT value yet reported for this gate length, and further improvements in both speed and off-state leakage We expected to be obtainable.

22

Claims

A field effect transistor (FET) of the kind comprising biasing means for depressing the intrinsic contribution to the charge carrier concentration in an intrinsic region (106) thereof, characterised in that the FET (100) includes means for defining a channel extending between a source region (110) and a drain region (112) with any intervening departure from channel straightness being not more than 50 run in extent, as appropriate to enable a high value of current gain cut-off frequency to be obtained.

2. An FET according to Claim 1 characterised in that any departure from channel straightness is not more than 5 run in extent.

3. An FET according to Claim 1 or 2 characterised in that it is an enhancement mode NfiSFET (100).

4. An FET according to Claim 1, 2 or 3 characterised in that it incorporates source and drain regions (110, 112) which are heavily n-type.

An FET according to any preceding claim characterised in that the intrinsic region (106) is p-type and forms extracting contact means in combination with the source and drain regions (110, 112).

An FET according to any preceding claim characterised in that the intrinsic region (106) has an interface with a barrier region (104) itself having an interface with a base region (102), and wherein the intrinsic, barrier and base regions (106, 104, 102) are of like 23 conductivity type and the barrier region (104) is of relatively wider bandgap than the intrinsic and base regions (106, 102) and provides an excluding contact to the intrinsic region (106).

7. An FET according to Claim 6 characterised in that it includes a gate contact (116) insulated from and extending at least over that part of the intrinsic region (106) between the source and drain regions (110, 112) to define an enhancement channel therebetween in operation.

8. An FET according to Claim 6 or 7 characterised in that:- a) the base region (102) is of P' InSb and has an acceptor concentration of at least 5xl 017 CM-3; b) the barrier region (104) is of 2' Inj M.Sb with x in the range 0.05 to 0.25 and has 017 - an acceptor concentration of at least 5xl cm-3. the intrinsic region (106) is of z InSb with an acceptor concentration of less than than 5x1017 cm-3, preferably in the rangelx1015 CM-3 to 5X1016 CM-3; and d) the source and drain regions (110, 112) are of n' InSb with a dopant concentration of at least 5x 1 0 17 CM-3.

c) 9. A FET according to Claim 5, 6, 7 or 8 characterised in that the base, barrier and intrinsic regions (102, 104, 106) are successively disposed'in a layer structure, and the intrinsic region (106) has a substantially flat surface portion supporting a gate insulation layer (108) and a gate contact (116).

Z:1 10. An FET according to Claim 1 or 2 characterised in that it is a depletion mode MISFET (200) haAng an associated channel region (208).

24 11. An FET according to Claim 10 characterised in that it incorporates source and drain regions (212, 214) which are heavily doped outgrowths formed upon either the intrinsic region (206) or the channel region (208), the outgrowths defining therebetween a gate recess (222) accornmodating a gate contact (218).

12. An FET according to Claim 10 or 11 characterised in that the intrinsic region (206) is p- type and either itself or the channel region (208) forms extracting contact means with the source and drain regions (212, 214).

13.

An FET according to Claim 10, 11 or 12 characterised in that the intrinsic region (206) has an interface with a barrier region (204) which itself has an interface with a base region (102), and wherein the intrinsic, barrier and base regions (206, 204, 202) are of like conductivity type and the barrier region (204) is of relatively wider bandgap than the intrinsic and base regions (206, 202) and provides an excluding contact to the intrinsic region (206).

An FET according to Claim 13 characterised in that:- a) the base region (102) is of p' InSb and has an acceptor concentration of at least 5xl 017 CM-3; b) the barrier region (104) is of p' Inj,Al,,Sb with x in the range 0.05 to 0.25 and has an acceptor concentration of at least 5xl 017 tM-3; c) the intrinsic region (106) is of 7r InSb with an acceptor concentration of less than than 5x 1()17 CM-3, preferably in the rangelx101' crn7' to 5x10'6 cm"; and d) the source and drain regions (110, 112) are of n' InSb with a donor concentration of at least 5x 1017 CM-3.

A FET according to Claim 13 or 14 characterised in that the intrinsic region (206) supports a channel region (208), the base, barrier, intrinsic and channel regions (202, 204, 206, 208) are successively disposed in a layer structure, the source and drain regions (212, 214) are grown upon the channel region (208) and the channel region (208) has a substantially flat surface portion supporting a gate insulation layer (210) and a gate contact (218).

16. An FET according to Claim 15 characterised in that the source and drain regions (212, 214) define therebetween a gate recess (222), the channel region (208) has a surface portion at an end of the recess (222) supporting a gate insulation layer (208) and a gate contact (210).

17. An FET according to Claim 10, 11 or 12 characterised in that the channel region lies between parts of the intrinsic region and the latter forms extracting contact means in combination with the source and drain regions (212, 214).

18. An FET according to Claim 17 characterised in that the base, barrier and intrinsic regions (202, 204, 206) are successively disposed in a layer structure, the intrinsic region (206) contains the channel region (208), the source and drain regions (212, 214) are supported by the intrinsic region (206) and define therebetween a gate recess (222), and the intrinsic reeion (206) has a surface portion at an end of the recess (222) supporting a gate insullation layer (208) and a gate contact (210).

19. An FET according to any preceding claim characterised in that the biasing means for deoressing the intrinsic contribution to the charge carder concentration in the intrinsic region (106, 206) is arranged to bias the FET (100, 200) at a point of infinite differential imDedance where the variation of gate threshold voltage with substrate bias voltage variations is minimised.

26 20. A method of making an FET of the kind comprising biasing means for depressing the intrinsic contribution to the charge carrier concentration in an intrinsic region (106) thereof, characterised in that the method includes defining a channel extending between a source region (110) and a drain region (112) such that any intervening departure ftom channel straightness is not more than 50 mn in extent, as appropriate to enable a high value of current gain cut-off frequency to be obtained.

21. A method of making an FET according to Claim 20 characterised in that any departure from channel straightness is not more than 5 ilm in extent.