Small-signal distributed FET model consistent with device scaling zyxwvuts A. zyxwvutsrqpo Cidronali, G. Collodi, G. Vannini and A. Santarelli A distributed modelling approach for micro- and millimetre-wave FETs is presented. Model identification is directly carried out on the bases of S-parameter measurements and electromagnetic analysis of the device layout, without requiring cumbersome optiinisation techniques. Experimental results confirm that the model is consistenl with device scaling. zyxwvutsrqp Itztruducnc.tioiz: The development of high performance monolithic microwave integrated circuits (MMICs) requires global design procedures where not only the values of passive components, but also the active device geometry (e.g. number of fingers and gate width) represent available design parameters. In this context, robust scaling procedures for FET models are of key importance. In this Letter, a new approach to FET model scaling for MMIC design is proposed which is based on an empirical distributed modelling approach [I zyxwvutsrqponm - 31. In particular, model identification is carried out by means of accurate electromagnetic (EM) simulation and scattering parameters measured for a limited number of FET structures. On this basis, a characterisation of the ‘active area’ of the device is obtained which is consistent with simple scaling rules. zyxwvut [ CQ?? Fig. 1 Structure of distributcd niodel zyxwvutsrq Empiricul distributed model: The electron device is assumed to con- sist of an ‘extrinsic passive structure’ connected with a finite, suit- able number of elementary ‘intrinsic active slices’ as shown in Fig. 1 for a two-finger FET where a single active slice per finger has been considered. As far as model identification is concemed, the extrin- sic structure is characterised through its scattering matrix zyxwvutsrq S (see Fig. I) computed by means of EM simulation. This kind of analy- sis enables device geometry and material characteristics to be taken into account, for any given device structure and size, by means of a multiport S-matrix ‘distributed’ description. Thus, since electromagnetic propagation and coupling effects are accounted for by the passive structure, all the intrinsic devices are described by the same scattering matrix BA, which can be identi- fied once the scattering matrix _a of the electron device has been measured. This identification procedure is well justitied by the experimental results provided in the following. The identification of the 3 x 3 matrix BA, which characterises the active slice, will be described for a two-finger FET where a sin- gle active slice per finger is considered as shown in Fig. 1. This choice, which does not limit the validity of the approach, has the advantage of simplifying the mathematical development and has been found suitable for-FET modelling up to 5OGHz. In particu- lar, by denoting with S the 5 x 5 matrix obtained from the 8 x 8 scattering matrix S of the extrinsic part, after the application of symmetry conditions deriving from the FET structure, it can be shown that with ?L=l,z,,=l ,.., 3 = zyxwvut sz=1,2;~=3,..,5 where the port indexes are defined according to Fig. 1. In eqn. 1, _I is the 3 x 3 identity matrix while _a is the 2 x 2 measured scattering matrix of the electron device. It is worth noting that model identi- fication does not require either parameter optimisation or complex measurements. Scciling ritlrs zyxwvuts jbr rirtive slices: Experinlentdl results show that the FET static currents and differential conductances deviate from a simple proportional relationship with device width. As an exam- ple, the saturated drain current and the static transconductance are plotted in Fig. 2 as functions of the total gate width for differ- ent GEC-Marconi MESFETs of the same wafer die (so that proc- ess dispersion is practically negligible). It is evident that although linear behaviour with gate width can be reasonably assumed, this is not purely proportional, as the linear regression does not pass through zero. A possible explanation could be related to the fact that the FET structure is not ‘geometrically homogeneous’ along the finger width, due to the gate feed and termination structures. These ‘border’ effects, which become more relevant at high fre- quencies, must be accounted for in the scaling rules. This can be done by introducing, under the reasonable assumption that border effects are independent of gate width, a correction term in the value of all the static and dynamic parameters. In particular, an equivalent admittance matrix Y,, associated to the active slice width can be defined according to the following scaling rule YAS,, (LJ, T,t‘~s) = (LJ) . II’As + Ci, (LJ) i, j = 1, ..., 3 (2) C(o) being a width-independent fraction of the equivalent admit- tance matrix, which accounts for non-ideal, border-like effects. (I) E 6-01 W,pm & Fig. 2 Linear regressions of’ mcwsnred I<,,,, and stutic trunsconductunce Gm against total gute periphery .~ Gtn Id,, . b5131 Fig. 3 Meusured und simukitrd S,, und S22,for three d$&vnt FETs H 2 x 150p 2 X 75pi zyxwvut 0 2 X 25pm Experimental results: Different GEC-Marconi MESFET structures were measured and simulated to validate the proposed approach. More precisely, the S matrices of the extrinsic passive structures were computed, on the basis of foundry process parameters and device GDSlI fdes, using the emTM Sonnet electromagnetic simuka- tor. The scattering parameters of the electron devices were meas- ured directly on-wafer up to a frequency of SOGHz using an ELECTRONICS LETTERS 4th March 1999 Vol. 35 No. 5 37 1 Authorized licensed use limited to: Universita degli Studi di Firenze. Downloaded on May 21,2010 at 09:12:16 UTC from IEEE Xplore. Restrictions apply.