RF Performance of Si / SiGe MODFETs : A Simulation Study

The microwave performance potential of Si/SiGe pseudomorphic MODFETs are studied, in comparison to state of the art InGaAs pseudomorphic HEMTs. Both devices have equivalent structures corresponding to a physical HEMT used for calibration. We use an RF analysis technique based on transient Monte Carlo simulations to estimate the intrinsic noise figures, the RF figures of merit fv and fmax, and the effect of contact and gate resistances. Both devices exhibit velocity overshoot below the gate region. It is shown that the difference in noise figures and fT values can be mainly attributed to differences in device channel velocity, fmax exhibits a strong dependence on device contact resistance, eroding some of the performance advantage of the pseudomorphic HEMT.


INTRODUCTION
Recent theoretical and experimental studies show that low field mobility and velocity overshoot are enhanced in Si layers grown pseudomorphically on relaxed SiGe substrates.Induced strain breaks the six-fold degeneracy of the Si conduction band, resulting in an .improvedband offset for the two conduction valleys whose transverse effective mass is in the plane of the heterojunction.This increases the in plane effective mobility and reduces inter- valley scattering in the Si layer [1].Modulation   doped field effect transistors (MODFETs) based on this material system have been demonstrated, and show significant potential for RF applications [2,3].Although measured mobilities are lower than those reported in optimal III-V based devices, the compatibility between SiGe and conventional Si processing technology makes such MODFETs attractive for Si MMICs design and microwave signal processing applications inte- grated on conventional Si chips.Therefore, comparison of strained Si channel devices with well established members of the III-V family will S. ROY et al.
provide greater insight into their potential performance and limitations.
As part of this goal, we here study the micro- wave performance potential of a Si/SiGe pseudo- morphic MODFET in comparison with a state of the art InGaAs channel pseudomorphic HEMT, using an RF analysis technique based on transient Monte Carlo simulations.

DEVICE STRUCTURES AND PERFORMANCE ANALYSIS
To allow a fair comparison similar device layer structures are considered (Figs.a,b).Both devices have tS-doping separated by a 2.5 nm spacer from the channel, a T shape recess gate with 50nm recess offset and heavily doped (n-type 1018 cm-3) cap layers.An effective &doping of 5x 1012 cm -2 is considered in the both cases.Total gate-to-channel separation is 22 nm and the gate length is 0.12 gm.p-type background substrate doping of 10 TMcm -3   is considered.This corresponds to the dimensions of real pseudomorphic HEMT fabricated at the Glasgow Nanoelectronics Research Centre and used for validation and calibration of the RF Monte Carlo analysis technique (described in an accompanying paper [4]).We realise that for the Si/SiGe MODFET this may pose unresolved growth problems.Specifically we conjecture im- /30.0nm Vertical layer structure of (a) a 0.12 gm pseudomorphic HEMT and (b) a 0.12 gm pseudomorphic MODFET.Both devices have the same vertical dimensions.
provements in growth technology and low tem- perature processing to allow formation of a well defined As &doping supply layer.For simplicity of Monte Carlo modelling a uniform SiGe substrate is assumed, instead of a 'virtual substrate' with graded Ge concentration.The 'etch stop' region of the Si/SiGe device is also considered as (undoped) SiGe.Simple 1-D Poisson calculations indicate that if a strained Si 'etch stop' region is included, it will exhibit negligible parallel conductance at a d.c.bias of V -0.25 V. RF analysis of HEMT and MODFET perfor- mance is based on Monte Carlo simulation of device transient response.We follow Yamada's [1] treatment of the effect of the strain on the Si chanpel band structure, with conduction band splitting of AEc 0.67x (eV) and band gap Eg 1.11 0.74x (eV).However the six phonon scatter- ing model of Jacoboni [5] is implemented, instead of the four phonon model in [1].Acoustic and ionised impurity scattering modes are also in- cluded, with ionised impurity scattering calculated from the Brooks-Herring model [6].Bulk velocity- field characteristics obtained from the model are in agreement with experimental unstrained Si data at 77 K and 300 K and previous results for strained Si [1,7].The transient Monte Carlo simulations begin by following 5 104 superparticles for 2 ps settling time at d.c.bias, then a further 2.02.6 ps during which device statistics are recorded at fs intervals.A step change A V 0.2 V, or A VD 0.3V is then consecutively applied, and the transient response measured for a further 2 ps.It is found that structure and doping dependant THz oscillations in device drain current (possibly due to plasma oscillations in the channel, or in the heavily doped cap layer) may mask the detailed form of the transients, and so a number oftraces are averaged to define the response.Complex y-parameters are derived by Fourier transforming these terminal current transients, and used to extract the small signal equivalent circuit, intrinsic noise figures, and estimate the RF performance figures of merit fT (cutoff frequency) and fma (maximum frequency of oscillation).Finally, the small signal equivalent circuit is augmented by the addition of external impedance, and thus the effect of contact and gate resistance on the 'real' device operation estimated.
A detailed description of the analysis process is given in [4].

RESULTS AND DISCUSSION
The average velocity in the channel of both devices is compared in Figure 2 for a d.c bias of VD 1.5 V and V -0.25 Vo In the pseudomorphic HEMT this corresponds to the region of maximum transconductance, while in the pseudomorphic MODFET it is the region of maximum transcon- ductance achieved while constraining parallel conduction in the -doped region to less than 10% of total conduction.The gate extends from x= -0.12 0.0 tm.Both devices show distinct overshoot below the gate region.Peak velocity in the pseudomorphic MODFET, at the drain end of the strained Si channel, approaches twice the saturation velocity of bulk unstrained Si.The ratio between peak velocity in the MODFET compared to that of the pseudomorphic HEMT InGaAs channel is 2.8.
Figures 3-5 and Table I characterise the RF performance of the two devices.The intrinsic noise figures NF (in decibels) are shown in Figure 3. From the definition ofNF we postulate that the significant difference between the devices will be in their transconductance gmo Y21-All other terms y- parameters and current fluctuations < 12> we therefore represent by an approximate constant C.

l+
(1) NF-+ 1-2 < 512g > gmo From Figure 3 and equation ( 1) the ratio of transconductances gr.tEMr/gMoirEr is indeed found to be approximately constant below 60 GHz, with value 2.6.The transconductance values directly obtained from the Monte Carlo simula- tions are gIJEMT 87 and goozzT 31, with ratio 2.8.It can be concluded that the bulk of the x (tm)  Gate Resistance () FIGURE 5 Variation Offmax in a pseudomorphic MODFET as a function of gate and contact resistances.
difference in noise figures between the two devices can be attributed to the differences in transcon- ductance-and thus to differences in channel velocity.The ratio of cut-off frequencies Jfor the pseudo- morphic HEMT and MODFET is approximately 2.5 and independent of external resistance.This implies that the value offx is primarily governed by the channel velocity of the respective device.
However, the ratio Offmax for the pseudomorphic HEMT and MODFET is 3.9 for intrinsic devices, and drops to 2.5 when external contact resistances of 5f are applied.A more realistic inclusion of device parasitics in this case reduces the advan- tages of the pseudomorphic HEMT.Finally, Figures 4 and 5 detail the effect of varying contact resistance on both figures of merit for the pseudomorphic MODFET, and show clearly the strong dependence offmax on device parasitics.

CONCLUSIONS
The microwave performance potential of a Si/SiGe pseudomorphic MODFET was studied, in com- parison to a state of the art InGaAs channel pseudomorphic HEMT.We used an RF analysis technique based on transient Monte Carlo simula- tions, calibrated to a physical HEMT.The difference in device noise figures was attributed mainly to differences in device transconductance, and thus to differences in channel velocity.The cut-off frequency fx was also shown to be governed by channel velocity.However the max- imum frequency of oscillation Jx was seen to exhibit a strong dependence on device contact resistance, eroding some of the advantage of the pseudomorphic HEMT over the Si/SiGe MOD- FET.
nanometre-scale fabrication and its application to electronic and optoelectronic devices.He has over 100 publications on electron beam nanolithography, dry etching, short-gate III-V based transis- tors, quantum transport devices, the optical properties of quantum dots, and single electron devices.Latterly he has become involved with the issue of manufacturability of mm-wave circuits and the use of nanometre-scale fabrication techni- ques coupled with technology-based device simt- lations to forecast performance and yield with the minimum of process iterations.

FIGURE 2 FIGURE 3
FIGURE 2 Comparison between average velocity in the channel of 0.12tm pseudomorphic HEMT and MODFET.Vt 1.5V, with V6=-0.25Vcorresponding to maximum transconductance.

FIGURE 4
FIGURE 4 Variation offT in a pseudomorphic MODFET as a finction of source and drain contact resistance (Jr invariant to changes in gate resistance).

TABLE
CalculatedRF figures of merit for pseudomorphic HEMT and MODFET at bias Vr 1.5V, VG=-0.25V.Device width of 100 gm is assumed, fx, fmx calculated for both negligible external gate and drain contact resistance, and for more realistic contact resistances of 5f Table I lists the RF figures of merit extracted from the transient response of each device.