JP4428190B2 - How to create a high-frequency transistor model - Google Patents

Description

The present invention relates to the creation how high-frequency transistor model for representing the transistor unit of the high-frequency semiconductor circuit on a computer.

In recent years, in the fine CMOS technology, a characteristic in which the current gain cutoff frequency ft exceeds 100 GHz has been obtained. For this reason, for example, CMOS transistors have started to be used for high-frequency communications such as wireless LAN and blue-tooth instead of conventional MESFETs and bipolar transistors using Group 3-5 semiconductors (GaAs, InP).
In addition to its low cost, the CMOS transistor has been improved in frequency characteristics by a high current gain cutoff frequency ft and a maximum oscillation frequency fmax. In addition, the CMOS transistor circuit has many advantages such as easy integration of a digital circuit and coexistence of an analog circuit and the utilization of SOC (System On a Chip) technology. For these reasons, CMOS transistors are employed in high frequency circuits with relatively low frequencies, and their use in circuits that handle higher frequencies is being studied.

  In the design of a high-frequency circuit for mobile communication, specifications for power consumption and noise are strict, and a circuit design technique based on highly accurate characteristic prediction is desired. Therefore, high-precision prediction of transistor characteristics at high frequency is a key factor for practical application of high-frequency CMOS semiconductor circuits.

As high-frequency transistor models for circuit simulation, for example, compact models such as BSIM3 ver3, MOS model 9, and EKV are known. These compact models have been developed for logic applications that presuppose clock operations in the MHz band, and need to be adapted to high frequency applications exceeding 10 GHz.
Currently, an inefficient method is generally used, in which measurement data (S-parameters, etc.) is prepared for each frequency or bias, and a look-up table is constructed by combining them to design a high-frequency circuit while always referring to it. Is. For this reason, the current compact model cannot reproduce characteristics other than the measured frequency and bias conditions. Therefore, an attempt is made to reproduce high-frequency transistor characteristics at high frequencies by adding an external circuit that expresses a parasitic transistor portion, for example, an external gate resistor, to the compact model (for example, non-patent literature). 1).

  A model that shows an operation when a model including an external circuit at a high frequency is viewed under a certain bias condition is called an equivalent circuit model. As an EDA (electronic design automation) tool that handles an equivalent circuit model in the frequency domain, for example, ADS (Advanced Design System: trade name) manufactured by Agilent is exemplified.

FIG. 15 is a cross-sectional view of the high-frequency transistor described in Non-Patent Document 1. FIG. 16 shows an equivalent circuit model including an external circuit that reproduces the operation of the high-frequency transistor shown in FIG.
In FIG. 16, the intrinsic transistor part Mi is a part provided by the compact model. The external circuit added to the internal transistor part Mi is connected to the substrate circuit between the intrinsic drain node di and the drain back bias node db. Junction capacitance Cj_db (capacitance of diode Dd), source / substrate junction capacitance Cj_sb (capacitance of diode Ds) connected between the intrinsic source node si and back bias node di, and back bias node di and back Three resistors connected between the bias terminal B, that is, a drain substrate resistance Rsub1, a source / drain substrate resistance Rsub2,3, and a source substrate resistance source substrate resistance Rsub4 are included. The external circuit includes a gate resistor Rg, a source resistor Rs, a drain resistor Rd, a gate-source overlap coupling capacitance Cgs_ov, and a gate-drain overlap coupling capacitance Cgd_ov.
Christian Enz, “An MOS Transistor Model for RF IC Design Valid in All Region of Operation”, IEEE Transaction on Microwave Theory And Techniques, vol. 50, No. 1, January 2002

However, since these compact models and models using external circuits simplify the phenomenon that the model actually occurs, the model itself may not be able to reproduce the phenomenon that actually occurs.
For example, the applicable frequency in the paper of Non-Patent Document 1 is 10 GHz, and accuracy is not obtained at a frequency higher than that because the model is too simplified.

  Particularly in a high-frequency operation region of 10 GHz or more, the gate terminal (voltage supply point of the gate electrode) G, the drain terminal (voltage supply point of the drain region) D, and the source terminal (voltage supply point of the source region) shown in FIGS. ) External circuit parameter values are set according to the wiring and contact structure (hereinafter simply referred to as layout) of the high-frequency semiconductor integrated circuit outside S and the back bias terminal (voltage supply point of the back bias impurity region) B It has not been. For this reason, when the layout is changed, the corresponding parameter value does not fluctuate, and the model is not a layout scalable model. This significantly limits the degree of freedom of layout change.

  The object of the present invention is to provide a method for determining the parameter value corresponding to the topology (configuration) of the equivalent circuit that reproduces the phenomenon actually occurring and the layout, and by using it, the source, drain, or gate When a layout such as a wiring or a contact structure is changed, a parameter value corresponding to the layout can be changed, and a new transistor model capable of layout change (Layout scalable) and a method for creating the same are provided.

The method for creating a high-frequency transistor model according to the present invention provides parameters that can be changed for a transistor unit that includes an intrinsic transistor portion and electrodes, contacts, and wirings that are connected to the intrinsic transistor portion and change according to the layout. When creating a transistor model, which is circuit data including, circuit parameters included in the device model of the intrinsic transistor unit from device simulation by a device simulation device using the device model and direct current measurement by a direct current measurement device. A first step for determining an intrinsic parameter of the intrinsic transistor part, a non-quasi-static parameter indicating a time delay of a carrier traveling in the channel of the intrinsic transistor part, and a resistance component depending on a skin effect And Respectively, a second step of obtaining a device simulation using device simulation apparatus, from the intrinsic transistor section, a source external terminal of said transistor unit, a drain external terminal, electrodes to the gate external terminal, or to the electrode With respect to the parasitic parameters of the wiring connected through the contacts, the resistance value and the inductance value that change depending on the layout are obtained from device simulation by a device simulation apparatus. At this time, the gate resistance is the same as the non-quasi-static parameter. A third step of using a combined resistance with the resistance component depending on the skin effect, and a fourth step of obtaining a capacitance value independent of the layout by device simulation using a device simulation device, A high-frequency transistor having an intrinsic parameter of the intrinsic circuit, an extrinsic parameter including a resistance component depending on the non-quasi-static parameter and the skin effect, and a parasitic parameter of resistance and inductor depending on the layout This is a model creation method.

Preferably, when at least one of the non-quasi-static parameter and the resistance component depending on the skin effect includes a frequency-dependent component that changes at a non-negligible rate as the operating frequency increases, in the device simulation of the third step The device simulation apparatus gives at least one of the non-quasi-static parameter and the resistance component depending on the skin effect as a function representing the frequency dependence.

The method for creating a high-frequency transistor model according to the present invention includes a layout parameter that changes in accordance with a change in the layout of the electrodes, wirings, and contacts of the transistor unit in the parasitic circuit. Therefore, the accuracy of simulation using the high-frequency transistor model Is expensive. Further, when the layout such as source, drain or gate wiring or contact structure is changed, the corresponding parameter value can be changed, and the layout can be changed (Layout scalable). Furthermore, when changing the layout , the layout parameters can be easily obtained from the layout itself, and it is not necessary to set parameters that do not depend on the layout and need not be changed.

  Embodiments of the present invention will be described below with reference to the drawings.

[First Embodiment]
In this embodiment, a configuration example of a high-frequency transistor model will be described.
FIG. 1 is a diagram showing a high-frequency transistor model according to the present embodiment, taking as an example a case where a source terminal and a back bias terminal are interconnected. Note that the source terminal and the back bias terminal need not be connected. In that case, the line directly connecting the source terminal S and the back bias terminal B shown in FIG. 1 is omitted.

  The transistor model of this illustrated example includes an intrinsic transistor section Mi, an extrinsic circuit, and a parasitic circuit.

The intrinsic transistor part Mi is a compact model such as “BSIM3 ver3 (Berkeley Short Channel IGFET Model3 version 3)” and “BSIM4” developed at the University of California, Berkeley, “MOS model 9 (trade name)” provided by Philips Semiconductor, This is a model of the active operation part of a transistor provided by “EKV” developed by EPFL (Electronics Laboratories, Swiss Federal Institute of Technology) in Switzerland.
Alternatively, an equivalent circuit model having parameters such as mutual conductance gm, drain conductance gds, intrinsic gate / drain capacitance Cgd_int, intrinsic gate / source capacitance Cgs_int, intrinsic gate / substrate capacitance Cgb_int, etc. Can also be replaced.

  The intrinsic transistor portion Mi includes a gate contact (hereinafter referred to as an intrinsic gate node) gi, a drain contact (hereinafter referred to as an intrinsic drain node) di, a source contact (hereinafter referred to as an intrinsic source node) si, and a back bias. It has four internal contacts called contacts (hereinafter referred to as intrinsic back bias nodes).

In the portion surrounded by a broken line in FIG. 1, the portion excluding the intrinsic transistor portion Mi indicates an extrinsic circuit.
The external circuit is connected to the intrinsic gate node gi, the intrinsic drain node di, the intrinsic source node si, and the intrinsic back bias node bi of the intrinsic transistor unit Mi, and is inherently included in the transistor. A network of parasitic elements is formed. More specifically, the external circuit includes a passive parasitic element that constitutes a substrate circuit of the high-frequency transistor and a passive parasitic element that exhibits a non-quasi static effect (NQS effect).

The external circuit has a drain-substrate junction capacitance Cj_db (capacitance of the diode Dd), a source-substrate junction capacitance Cj_sb (capacitance of the diode Ds), and an intrinsic back bias node bi and back as passive parasitic elements of the substrate circuit. Four substrate resistors connected between the bias terminal B, that is, a drain substrate resistor Rsub1, a source / drain substrate resistor Rsub2 and Rsub3, and a source substrate resistor Rsub4 are included. Hereinafter, these four resistors are simply referred to as “substrate resistance”. Substrate capacitances Csub1, Csub2, Csub3 and Csub4 are connected in parallel with the corresponding substrate resistance among the four substrate resistances Rsub1 to Rsub4 as shown in FIG.
The external circuit includes a source resistor Rs, a drain resistor Rd, a gate-source overlap coupling capacitance Cgs_ov, and a gate-drain overlap coupling capacitance Cgd_ov.
Furthermore, the external circuit includes an NQS gate resistance Rg_NQS as a passive parasitic element representing the NQS effect.

The parasitic circuit is a circuit newly added in the present embodiment, and the electrode and wiring structure of the high-frequency transistor and the contact structure are arranged as passive parasitic elements further outside the external circuit, that is, in the transistor unit of the high-frequency semiconductor circuit. This is an equivalent circuit model approximated by
The parasitic circuit is shown in FIG. 1 by a portion other than the portion surrounded by a broken line.

Specifically, the passive elements included in the parasitic circuit are connected in series with the NQS gate resistor Rg_NQS between the intrinsic gate node gi and the external gate terminal G0 of the transistor unit, as shown in FIG. The gate resistance Rg0 and the inductance component Lg_layout of the gate wiring, and the layout component of the source wiring resistance connected in series with the source resistance Rs between the intrinsic source node si and the external source terminal S0 of the transistor unit. Rs_layout and the inductance component Ls_layout of the source wiring, and the layout component Rd_layout and the drain of the drain wiring resistance connected in series with the drain resistance Rd between the intrinsic drain node di and the external drain terminal D0 of the transistor unit. Wiring Including the inductance component Ld_layout.
The parasitic circuit is connected in parallel with the gate wiring / substrate coupling capacitance Cc_gb and the gate coupling substrate resistance Rsub_gb connected in series between the gate wiring and the substrate reference potential, and the gate coupling substrate resistance Rsub_gb. Gate coupling substrate capacitance Csub_gb. Similarly, the parasitic circuit is connected in parallel with the drain wiring / substrate coupling capacitance Cc_db and the drain coupling substrate resistance Rsub_db connected in series between the drain wiring and the substrate reference potential, and the drain coupling substrate resistance Rsub_db. Drain-coupled substrate capacitance Csub_db. Further, the parasitic circuit includes a gate / drain wiring coupling capacitor Cc_gd between the gate wiring and the drain wiring and a gate / source wiring coupling capacitor Cc_gs between the gate wiring and the source wiring.

  Here, the gate resistance Rg0 is a sum of a layout component Rg_layout of the gate resistance and an electromagnetic field effect component Rg_em due to a skin effect or the like.

  Here, since the NQS effect and the NQS gate resistance Rg_NQS described above need to be explained a little, they will be described below.

  In the paper of Non-Patent Document 1, a phenomenon in which velocity saturation of electrons occurs significantly in the case of an intra-channel traveling carrier, that is, an N-type channel and a P-type channel is reported. Along with miniaturization, the gate oxide film becomes thinner and the vertical electric field of the gate voltage becomes sufficiently strong, so that velocity saturation occurs in carrier movement, and a phenomenon in which the delay time of carrier movement cannot be ignored at high frequencies occurs. . This delay in carrier movement caused by the velocity saturation is called the NQS effect.

FIG. 2 shows the coupling between the gate electrode and the channel and the channel resistance at high frequencies as distributed constants. FIG. 3 shows a schematic configuration when the intrinsic transistor portion Mi is an equivalent circuit model.
In the high frequency region, as shown in FIG. 2, it is necessary to express the coupling capacitance (mainly oxide film capacitance Cox) between the gate electrode and the channel and the channel resistance Rch as distribution constants. In this case, due to the time delay until the carrier supplied from the source side reaches the drain side, for example, the time for charging the channel and gate capacitance Cox differs depending on the location, which affects the high-frequency operation of the device. Therefore, the longer the gate length, the larger the amount of electron delay, which affects the device operation as an NQS effect.

  Further, the influence of the NQS effect appears on several parameters such as the channel resistance Rch, the mutual conductance gm, and the output resistance Rds (= 1 / gds) shown in FIG. 3 as parameters of the equivalent circuit of the intrinsic transistor portion Mi. In the conventional compact model (BSIM3ver3, EKV, etc.), these parameter values are constant even when the frequency becomes high. Therefore, the behavior of an actual high-frequency transistor exceeds the parameter value at which the NQS effect appears. It deviates from the behavior at the low frequency specified by

  In so-called compact models (BSIM3 ver3, EKV, etc.), there are an NQ mode that does not incorporate the NQS effect and an NQS mode that incorporates the NQS effect, and generally reproduces low-frequency operation in the NQ mode.

In the NQ mode, a parameter value corresponding to the actual physical quantity is set in the conventional equivalent circuit shown in FIG. 15, the S parameter of the equivalent circuit is obtained by simulation, and the S parameter of the actual transistor corresponding to the equivalent circuit is measured. The two were compared.
4A to 4D are graphs of S parameters for comparing the actual measurement value and the simulation value (denoted as “Sim.”) At this time. As shown in these figures, in the simulation based on the conventional equivalent circuit shown in FIG. 15, it can be seen that the characteristic is shifted at a high frequency of about 10 GHz or more. This is the amount of deviation that occurs because the NQS effect is not incorporated.

In the conventional compact model, in order to reproduce the NQS effect, instead of the channel resistance Rch (see FIG. 3) as a lumped constant, a gate resistance (such as a sheet resistance of the gate electrode) Rg as a parasitic element with respect to the gate on the input side. In addition to the above, it has been proposed to add distributed channel resistance viewed from the gate. Since this distributed channel resistance was proposed by elmore, it is generally called “elmore delay” or “elmore resistance”.
In the present embodiment, as shown in FIG. 1, the distributed resistance (NQS gate resistance) Rg_NQS component as the elmore resistance is included in the gate resistance for the above reason.

  However, simply adding this NQS gate resistance Rg_NQS to the gate resistance representing the sheet resistance and gate contact resistance of the gate electrode cannot correct the deviation accurately by comparing the S parameter between the simulation and the actual measurement.

For example, FIG. 5 shows the frequency characteristics of the gate resistance value extracted from the actual S parameter.
FIG. 5 shows that the gate resistance Rg_spara extracted from the S parameter increases as the frequency increases, and this may be due to an increase in resistance due to the skin effect due to the electromagnetic field effect. For this reason, it is necessary to consider the influence of the electromagnetic field effect in order to accurately reproduce the S parameter.

  Therefore, in the present embodiment, as shown in FIG. 1, the gate resistance Rg0 in the parasitic circuit is determined by the layout component Rg_layout of the gate resistance calculated from the layout including the sheet resistance and gate contact resistance of the gate electrode and the gate wiring, It is expressed as a sum with an electromagnetic field effect component Rg_em by a skin effect or the like. A specific example of this resistance component separation method will be described in another embodiment.

  In an actual device, the electromagnetic field effect component Rg_em due to the NQS gate resistance Rg_NQS and the skin effect increases with the operating frequency. Therefore, in the present embodiment, at least one parameter of the NQS gate resistance Rg_NQS and the electromagnetic field effect component Rg_em due to the skin effect or the like, preferably both parameters have frequency dependency. As a method of giving frequency dependence, the parameter may be expressed as a function of frequency, or the optimum value of each parameter is stored in advance in a memory as a table for each frequency, and the use frequency is determined. In response to this, the optimum value of each parameter may be read, and the NQS gate resistance Rg_NQS and / or the value of the electromagnetic field effect component Rg_em due to a skin effect or the like may be automatically set.

Next, the output resistance will be described.
In an actual device, the above-described NQS effect causes a time delay of carriers reaching the drain, which is accompanied by an increase in the output resistance Rds. However, the conventional compact model does not reproduce the NQS phenomenon on the output side (drain side) because the output resistance Rds shown in FIG. 3 is constant regardless of the frequency.
Conventionally, as a method for correcting the deviation of the high frequency characteristics accompanying the increase in the output resistance Rds, simulation is performed, and the substrate resistance value is set so that the output reflection coefficient S22 when the output side is terminated with the characteristic impedance matches the ideal curve. Was fitting. For this reason, it has been necessary to repeat the simulation and the parameter change several times.

In the circuit diagram shown in FIG. 6, parameters necessary for suppressing the influence of an increase in the output resistance Rds at a high frequency by the characteristic adjustment by the conventional fitting are extracted and shown. FIG. 7 shows the locus of the S parameter (output reflection coefficient) S22 in the circuit shown in FIG.
The circuit shown in FIG. 6 has an output resistance in the intrinsic transistor part Mi connected between the drain terminal D and the source terminal S (the same potential as the back bias terminal B) as a parameter affecting the output reflection coefficient S22. In addition to Rds, a substrate circuit portion (external circuit portion) connected in parallel with the output resistor Rds is shown.
As shown in FIG. 6, the substrate circuit portion includes a drain junction capacitance Cj_db and a substrate resistance Rsub1 connected in series between the drain terminal D and the source terminal S, and a drain junction capacitance Cj_db and a substrate resistance Rsub1. Including a substrate resistance Rsub2 and a source junction capacitance Cj_sb connected in series between the connection midpoint of the source and the source terminal S.
Here, since the source resistance Rs and the drain resistance Rd shown in FIG. 1 are negligibly small compared to the output resistance Rds, they are omitted in FIG. The substrate resistances Rsub3 and Rsub4 on the source side are omitted because the contribution to the output reflection coefficient S22 is relatively small. Also, the substrate capacitances Csub1 to Csub4 parallel to each substrate resistance are omitted for the sake of simplicity.

When the output resistor Rds and the substrate circuit portion parallel to the output resistor Rds are represented by the configuration shown in FIG. 6, the behavior of the output reflection coefficient S22 on the Smith chart is as shown in FIG. In FIG. 7, since the output resistance Rds is constant regardless of the frequency, the impedance of the circuit shown in FIG. 6 is not the case where the substrate resistance Rsub2 is taken into consideration, that is, “Rds // (Rsub1 // Rsub2)”. And “Rds // Rsub1”, it can be seen that there is a deviation (impedance is low) from the iso-resistance surface (ideal curve).
This is the reason why the actual measurement value and the simulation value are different from each other in FIG.

In order to correct this deviation, when a conventional transistor model is used, a method of fitting a substrate resistance value with reference to a simulation result has been adopted.
FIG. 8 is a table showing a comparison between the substrate resistance value (measured value) obtained by calculation from the layout and the substrate resistance value extracted after fitting the S parameter.
In FIG. 8, the value 250 [Ω] of the resistance Rsub1 extracted from the actually measured output reflection coefficient S22 is larger than the value 50 [Ω] of the substrate resistance Rsub1 measured from the layout. This does not reflect the effect of increasing the output resistance Rds as NQS at a high frequency because the compact model fixes the output resistance Rds. Therefore, when the S parameter is fitted, the total power of the output reflection coefficient S22 is adjusted. As described above, the substrate resistance Rsub increases, and this can be explained by reducing the power consumption on the substrate side.

Therefore, in the present embodiment, the frequency dependency is given to the output resistance Rds in the intrinsic transistor portion Mi. As a method of giving frequency dependence, the parameter (output resistance Rds) may be expressed as a function of frequency, or the optimum value of the output resistance Rds for each frequency is stored in advance as a table in a memory. When the use frequency is determined, an optimum value of the output resistance Rds may be read out and the output resistance parameter may be automatically set accordingly.
Further, all or some of the substrate resistances Rsub1 to Rsub4, for example, the substrate resistance Rsub1, are set so as to cancel the influence of the increase in the output resistance Rds depending on the frequency while keeping the output resistance Rds at a constant value such as a default value. And Rsub2 may have frequency dependency. Even in this case, as a method of giving frequency dependence, the parameter (substrate resistance) may be expressed as a function of frequency, or the optimum value of substrate resistance for each frequency is stored in advance as a table in a memory. In addition, when the use frequency is determined, an optimum value of the substrate resistance may be read in accordance with the determined frequency and the substrate resistance parameter may be automatically set. In addition to the substrate resistance, the substrate capacitance can also be given frequency dependence by the same method as described above.

  In the case where the substrate resistance is made to have frequency dependency, it is not described in FIG. 1 because it is complicated. However, each of the substrate resistances Rsub1 to Rsub4 separates the layout component from the frequency dependent component (NQS component). It is desirable to have a parameter structure. That is, the drain substrate resistance Rsub1 is indicated by the sum of the layout component Rsub1_layout and the NQS component Rsub1_NQS, the source / drain substrate resistance Rsub2 is indicated by the sum of the layout component Rsub2_layout and the NQS component Rsub2_NQS, and the source / drain substrate resistance Rsub3 is determined by the layout. The sum of the component Rsub3_layout and the NQS component Rsub3_NQS is indicated, and the source substrate resistance Rsub4 is indicated by the sum of the layout component Rsub4_layout and the NQS component Rsub4_NQS. A specific example of this resistance component separation method will be described in another embodiment.

  In the present embodiment, the NQS phenomenon on the output side is reproduced by varying each parameter value of the output resistance Rds and / or the substrate resistances Rsub1 to Rsub4 of the external circuit, thereby causing a phenomenon at a high frequency. That is, the influence of the increase in the output resistance Rds (= 1 / gds) at the high frequency on the high frequency characteristics is suppressed.

  In the present embodiment, by giving the output resistance Rds and / or substrate resistances Rsub1 to Rsub4 frequency dependence, it is only necessary to provide information on the operating frequency without performing parameter fitting while referring to the simulation result. It is an optimal parameter value. As a result, the transistor model of the present embodiment is actually measured with respect to the measured value of the S parameter of the actual device, particularly in the input reflection coefficient S11, the forward transmission coefficient S21, and the output reflection coefficient S22 mainly determined by the output resistance Rds. It is in good agreement with the value.

Note that the transistor model illustrated in FIG. 1 is an example of a desirable configuration.
As described above, the intrinsic transistor portion Mi may be a compact model or an equivalent circuit model.

In the present embodiment, the types and combinations of layout components included in the transistor model are arbitrary. Accordingly, it is sufficient that at least one of the various layout components described so far is included in the high-frequency transistor model.
In such a layout component shown in FIG. 1, the layout component Rg_layout of the gate resistance, the coupling capacitances Cc_gb and Cc_db of the wiring and the substrate, the substrate resistances Rsub_gb and Rsub_db and the substrate capacitances Csub_gb and Csub_db connected to the coupling capacitance, There are a coupling capacitance Cc_gd, a gate-source wiring coupling capacitance Cc_gs, and wiring inductor components Lg_layout, Ls_layout and Ld_layout. Further, layout components not directly shown in FIG. 1 include layout components Rsub1_layout to Rsub4_layout of substrate resistances Rsub1 to Rsub4 in the external circuit.
Further, when the source terminal S and the back bias terminal B are not short-circuited, the source-substrate coupling capacitance, the substrate resistance and the substrate capacitance, and the source wiring-drain wiring coupling capacitance, as well as the drain side, It is necessary to add to this layout component.
Of the substrate resistance and substrate capacitance constituting the substrate circuit of the external circuit, in particular, the substrate resistances Rsub1 and Rsub4, and the substrate capacitances Csub1 and Csub4 are positions of the application point of the back bias voltage in addition to the original layout of the transistor. Therefore, the layout parameter may be included in the parasitic circuit.

The external circuit may have a different configuration from that of FIG. 1 depending on the semiconductor device structure.
For example, in an SOI (silicon-on-insulator) transistor, the drain junction capacitance Cj_db and the source junction capacitance Cj_sb shown in FIG. 1 can be replaced with an insulating film capacitance such as a so-called box oxide film. In general, since the insulating film capacitance is considerably larger than the junction capacitance, it is possible to omit or simplify the substrate circuit constituted by the substrate resistances Rsub1 to Rsub4 and the substrate capacitances Csub1 to Csub4. Although the semiconductor substrate is in an electrically floating state or is electrically fixed at a constant voltage, the semiconductor substrate (back bias terminal B) is not normally connected to the source terminal S as shown in FIG.
On the other hand, when the influence on the semiconductor substrate side can be ignored, the substrate circuit shown in FIG. 1 can be regarded as an equivalent circuit in the SOI body region. In this case, the values of substrate resistance and substrate capacitance are greatly different from those of normal high-frequency transistors, and the body region is electrically floating depending on whether the operation of the SOI type transistor is a partial depletion type or a full depletion type. Depending on whether it is fixed or fixed, the configuration of the substrate circuit is also different.
Further, the two overlapping coupling capacitors Cgs_ov and Cgd_ov shown in FIG. 1 are included in the intrinsic transistor portion Mi (for example, included in the capacitors Cgd_int and Cgs_int in the equivalent circuit model shown in FIG. 3), and from the external circuit. May be omitted.

The high-frequency transistor model according to the present embodiment has the following advantages.
First, the high-frequency transistor model according to the present embodiment has a parasitic circuit including a parasitic component due to layout, and parameters that change to some extent depending on the layout are incorporated in the transistor model from the beginning. It is a high-frequency transistor model close to the device.

  Second, the high-frequency transistor model according to the present embodiment includes an NQS gate resistor Rg_NQS that gives an influence on the gate control of the time delay of the intra-channel traveling carrier to the gate resistance of the external circuit, and the parasitic circuit The gate resistance Rg0 includes an electromagnetic field component Rg_em as a component that gives influence on the gate control of an electromagnetic field effect such as a skin effect, and this can reproduce a phenomenon at a high frequency. It is a model.

  Thirdly, the high-frequency transistor model according to the present embodiment has components Rsub1_layout to Rsub4_layout in which each of the substrate resistances Rsub1 to Rsub4 of the external circuit or any necessary substrate resistance is obtained from the layout, and in the channel. It is shown as the sum of drain NQS components Rsub1_NQS to Rsub4_NQS that give influence on the output resistance Rds of the time delay of the traveling carrier. For this reason, the output resistance Rds can be changed with respect to the frequency in substantially the same manner as the actual device, and therefore, a high-frequency transistor model close to the actual device is obtained with higher accuracy.

  Fourth, at least one of the parameters representing the output resistance Rds in the intrinsic transistor part Mi, the substrate resistances Rsub1 to Rsub4 in the external circuit, or any necessary substrate resistance has frequency dependence. Therefore, the output reflection coefficient S22 has a frequency characteristic close to that of the actual device. Therefore, the high-frequency transistor model is more accurate and close to an actual device.

  As a comprehensive advantage by combining any one of the first to fourth advantages described above or some of them, a layout-scalable high-frequency transistor model is realized in the present embodiment. In other words, the high-frequency transistor model according to the present embodiment does not require parameters to be determined by the fitting method, or even if necessary, the parameters can be optimized with a slight change, so that the layout can be easily changed. It is.

[Second Embodiment]
An example of a method for creating a high-frequency transistor model will be described focusing on a method for determining (extracting and determining) parameter values.

  The high-frequency transistor model creation step includes a step of creating a model outline by creating an equivalent circuit described in the first embodiment, for example, as shown in FIG. 1, and a step of creating this outline (equivalent circuit). It is roughly divided into steps for determining each parameter value. In the present embodiment, the step of determining this parameter value will be mainly described.

FIG. 9 shows the main steps for determining parameter values.
In this figure, for convenience, the parameter determination flow of the intrinsic transistor part, the parameter determination flow of the external circuit, and the parameter determination flow of the parasitic circuit are shown separately. However, actual parameter determination is not limited to this, and it is efficient to perform parameter determination for each type of work such as measurement, calculation, and simulation. Further, the order of the steps is arbitrary as long as there is no contradiction in the parameter usage relationship. That is, as a matter of course, the parameter value used for calculation or the like must be determined prior to the calculation, and the order of the steps is arbitrary as long as this is observed.
Further, the specific parameter extraction method shown here is merely an example, and the present invention is not limited to this. Further, the type of simulator (trade name), ie, the device simulator “Medici”, the capacity simulator “SENECA”, the substrate simulator “substrate stream”, and “DESISS 3D Sim” are merely examples.

As shown in FIG. 9, the DC characteristics of the high-frequency transistor (for example, measurement of current-voltage (IV) characteristics) are measured in step ST1a for the intrinsic transistor portion, and the mutual conductance gm and drain conductance gds (that is, output) are measured. Resistance Rds) is obtained.
Further, the capacitance-voltage (C-V) measurement of the high-frequency transistor is performed to determine the intrinsic gate / drain capacitance Cgd_int.

In step ST4a, device simulation (D.Sim.) Of the high-frequency transistor is performed to determine the intrinsic gate / source capacitance Cgs_int, the intrinsic drain / source capacitance Cds_int, and the intrinsic gate / substrate capacitance Cgb_int. Here, for example, “Medici” is used.
Of the parameters of the intrinsic transistor section, the capacitance connected to the gate, that is, the intrinsic gate / drain capacitance Cgd_int, the intrinsic gate / source capacitance Cgs_int, and the intrinsic gate / substrate capacitance Cgb_int are a compact model or equivalent circuit model. Alternatively, a command value (op value) already prepared may be used.

Next, parameter determination for the external circuit will be described.
In step ST2b, calculation for extracting parasitic components is performed to obtain a drain resistance Rd, a source resistance Rs, a drain junction capacitance Cj_db, and a source junction capacitance Cj_sb.
The drain resistance Rd and the source resistance Rs can be obtained from the same calculation formula when the design of the diffusion layers (source region and drain region) on the source side and the drain side is symmetric. This calculation can be performed using the following equation (1), taking the drain resistance Rd as an example.

[Equation 1]
Rd = (W / Ld) · Rsheet_d
+ (W / Lext) ・ Rsheet_ext
+ Rcon_d / Ncon_d (1)

  Here, “W” is the width of the drain region and its extension, “Ld” is the effective length from the channel side end of the drain region to the contact, “Lext” is the length of the extension, and “Rsheet_d” is the drain The sheet resistance of the region, “Rsheet_ext” indicates the sheet resistance of the extension, “Rcon_d” indicates the contact resistance for one piece, and “Ncon_d” indicates the number of contacts (an integer of 1 or more).

  The drain junction capacitance Cj_db and the source junction capacitance Cj_sb can be obtained from the same calculation formula when the design of the diffusion layers (source region and drain region) on the source side and the drain side is symmetric. This calculation can be performed using the following equation (2) when the drain diffusion capacitance Cj_db is taken as an example.

[Equation 2]
Cj_db = Sarea / Cunit_area
+ Lperi ・ Cunit_peri… (2)

  Here, “Sarea” is the area of the area when the drain region is divided into the area part and the peripheral part (including the extension part) around the area part, “Lperi” is the width of the peripheral part, and “Cunit_area” is the area part. “Cunit_peri” indicates the junction capacity per unit width of the peripheral portion.

  In step ST3b shown in FIG. 9, the calculation for extracting the layout component is performed to obtain the layout component Rsuby_layout (y = 1, 2, 3, 4) of the substrate resistance. In this calculation, calculation is performed using an effective in-substrate distance between the source region and the drain region, an effective in-substrate distance from the source region or the drain region to the back bias feeding point, and the substrate sheet resistance. The layout component Rsuby_layout (y = 1, 2, 3, 4) of the substrate resistance can be obtained from the result of simulation (“Substrate strom” or “3D Sim. Of DESISS”).

In step ST4b, device simulation (D.Sim.) Of the high-frequency transistor is performed to obtain the NQS gate resistance Rg_NQS. Here, for example, “Medici” is used.
Further, a substrate simulation (S. Sim.) Is performed to determine the substrate resistance Rsuby and the substrate capacitance Csuby (y = 1, 2, 3, 4). Here, for example, “substrate stream” or “DESISS 3D Sim.” Is used. The substrate resistance Rsuby may be obtained from the measured output reflection coefficient S22.

In step ST5b, calculation for separating the NQS component from the substrate resistance is performed. The NQS component Rsuby_NQS (y = 1, 2, 3, 4) of the substrate resistance is obtained by subtracting the layout component Rsuby_layout from the substrate resistance Rsuby extracted from the measured value.
As described in the first embodiment, when making the output resistance Rds in the intrinsic transistor portion have frequency dependence, the steps ST5b and ST3b can be omitted.

Next, parameter determination for the parasitic circuit will be described.
In step ST1c, DC measurement of the high-frequency transistor is performed to determine the total gate resistance Rg_total. In this example, the total gate resistance Rg_total is obtained in advance because it is necessary for the calculation (step ST5c) for separating the electromagnetic field effect component of the gate resistance to be described later. If it can be obtained, this first step ST1c is unnecessary.

In step ST3c, a calculation for extracting a layout component is performed to obtain a layout component Rg_layout of the gate resistance, a layout component Rd_layout of the drain resistance, and a layout component Rs_layout of the source resistance.
The layout component Rg_layout of the gate resistance differs depending on how to make the gate contact. Taking as an example a case where the number of fingers of the gate electrode is M (M is an integer of 1 or more) and the gate wiring is one layer, the layout component Rg_layout of the gate resistance can be obtained by the following equation (3).

[Equation 3]
Rg_layout = k · (Lg / (M · Wfinger)) · Rsheet_gf
+ Rcon_gf / Ncon_gf
+ (L1mg / W1mf) · Rsheet_1mg) (3)

  Here, “k” is the number of values according to how to make the contact. For example, when the contact is made on one side of the gate finger portion, 1/3 is set, and when the contact is made on both sides, the value is set to 1/12. Good. “Lg” is the effective length of the gate finger, “L1 mg” is the effective length of the gate wiring, “Wfinger” is the width of the gate finger, “W1 mg” is the effective width of the gate wiring, “Rsheet_gf” "Is the sheet resistance of the gate finger (gate electrode)," Rsheet_1mg "is the sheet resistance of the gate wiring," Rcon_gf "is the contact resistance of one gate finger and gate wiring, and" Ncon_gf "is the number of contacts (1 or more Integer).

The layout component Rd_layout of the drain resistance and the layout component Rs_layout of the source resistance can also be calculated by substantially the same formula, although the value of the coefficient k is different from that of the gate resistance.
In the same step ST3c, an inductor component Lg_layout of the gate wiring, an inductor component Ld_layout of the drain wiring, and an inductor component Ls_layout of the source wiring are obtained. In addition to calculation, these may be fixed to specific values or may be obtained by simulation. In this example, when the increase in the inductor component of the gate wiring due to the high frequency operation is included in the electromagnetic field effect component Rg_em obtained in step ST5c described later, it is desirable to fix the inductor component of the gate wiring. Although not described in detail, the method of reflecting the increase in the inductor component of the wiring accompanying the high-frequency operation in the model can be similarly applied to the drain wiring and the source wiring.

In step ST4c, capacitance simulation (C.Sim.) Is performed to obtain a gate / drain wiring coupling capacitance Cc_gd, a gate / source wiring coupling capacitance Cc_gs, a gate wiring / substrate coupling capacitance Cc_Gb, and a drain wiring / substrate coupling capacitance Cc_db. . Here, for example, “SENECA” is used.
Further, a substrate simulation (S. Sim.) Is performed to obtain a gate coupled substrate resistance Rsub_gb and a gate coupled substrate capacitance Csub_gb, and a drain coupled substrate resistance Rsub_db and a drain coupled substrate capacitance Csub_db. Here, for example, “substrate stream” is used.

  In step ST5c, calculation for separating the electromagnetic field effect component from the total gate resistance is performed. The electromagnetic field effect component Rg_em of the gate resistance is obtained by subtracting the layout component Rg_layout and the NQS component Rg_NQS from the total gate resistance Rg_total.

  The parameter determination method shown in FIG. 9 includes a step of extracting layout components by calculation (for example, ST3b and ST3c), and more preferably, using the layout components, frequency-dependent components such as NQS components or electromagnetic field effect components are used. It is characterized by having a step (for example, ST5b, ST5c) that is separated by calculation. Since all of these can be obtained by simple calculation, in this embodiment, there is an advantage that a layout-scalable high-frequency transistor described in the first embodiment can be easily created.

  In FIG. 9, the substrate resistance Rsub is extracted from all the four substrate resistances Rsub1 to Rsub4. However, for example, only one substrate resistance Rsub1 or one to three substrate resistances such as substrate resistances Rsub1 and Rsub2 may be extracted.

  By setting the parameter values determined as described above to each parameter obtained in the model outline creation step, the high-frequency transistor can be accurately identified on a computer when the transistor characteristics are obtained at a desired high frequency. Creation of the model to represent is completed.

  Note that the basics of the model creation method described above can be similarly applied to the high-frequency transistor formed on the SOI substrate. However, it should be noted that the substrate circuit does not exist or the influence may be very small even if it exists. Therefore, the following method is desirable.

First, the output resistance Rds or the output reflection coefficient S22 is measured using a device simulator.
Next, an equivalent circuit of a signal transmitted to the semiconductor substrate through the box oxide film is configured. This equivalent circuit is connected to the semiconductor substrate via the output resistor Rds and the box oxide film, and device simulation is further performed. The difference between the simulation result when this equivalent circuit is connected and the device simulator result when the equivalent circuit is not connected is calculated as an NQS component representing the amount of charge delay. It is desirable to give this NQS component frequency dependence. As a method for providing a frequency-dependent component, a function value that increases with frequency may be used, or a value corresponding to the frequency may be used as a table.
Then, by putting other parameters into the equivalent circuit and causing circuit operation, transistor characteristics at a desired high frequency can be obtained. In this way, the characteristics of the high-frequency SOI transistor are calculated from the equivalent circuit.

[Third Embodiment]
The present embodiment relates to a model re-creation (update) method in the case of changing the layout after the high-frequency transistor model shown in the first embodiment is created by the method shown in the second embodiment, for example. .

FIG. 10 is a flowchart showing an outline of a method for re-creating (updating) a model accompanying a layout change.
In step ST11, each parameter is extracted from the first layout. For this parameter extraction, the method used in the second embodiment described with reference to FIG. 9 can be suitably used. As a result, the values of the parameters of the intrinsic transistor unit, the extrinsic circuit, and the parasitic circuit for the first layout are extracted or calculated.

In step ST12, in the high frequency transistor of the second layout to be changed, transistor characteristics at DC are measured, and parameters obtained therefrom are extracted.
Transistor characteristics at DC measured in step ST12 include mutual conductance gm, drain conductance gds (output resistance Rds), and the like.

In step ST13, various resistances and various capacitances are estimated from the second layout.
The layout components that can be estimated from the second layout include the layout component Rg_layout of the gate resistance, the layout component Cgs_layout of the gate / source capacitance, the layout component Cgd_layout of the gate / drain capacitance, the layout component Csb_layout of the source / substrate capacitance, There are a layout component Cdb_layout of the substrate capacitance, a layout component Cgb_layout of the gate / substrate capacitance, and layout components Rsub1_layout to Rsub4_layout of the substrate resistances Rsub1 to Rsub4.

Here, the layout component Cgs_layout of the gate / source capacitance, the layout component Cgd_layout of the gate / drain capacitance, the layout component Csb_layout of the source / substrate capacitance, the layout component Cdb_layout of the drain / substrate capacitance, and the layout component Cgb_layout of the gate / substrate capacitance are described above. Although these are new parameters that are not used in the second embodiment, these are extracted only for the layout component of the overall capacity shown in the second embodiment.
That is, the layout component Cgs_layout of the gate-source capacitance is a total of the intrinsic gate-source capacitance Cgs_int, gate-source overlap capacitance Cgs_ov, and gate-source wiring coupling capacitance Cc_gs shown in the second embodiment. In this example, only the layout components are extracted from the capacitance, and the gate and source are assumed on the assumption that the source wiring material and dimensions (thickness and width) and the wiring routing are the same in the first and second layouts. The capacitance layout component Cgs_layout can be estimated purely from the layout.

The same applies to other new layout components.
The layout component Cgd_layout of the gate / drain capacitance is shown by extracting only the layout component from the total capacitance indicated by the sum of the intrinsic gate / drain capacitance Cgd_int and the gate / drain wiring coupling capacitance Cc_gd shown in the second embodiment. Is.
The source / substrate capacitance layout component Csb_layout when the source terminal S and the back bias terminal B are not connected is indicated by the sum of the source / substrate junction capacitance Cj_sb and a part of the substrate capacitance shown in the second embodiment. Only the layout components are extracted from the total capacity and shown.
The layout component Cdb_layout of the drain / substrate capacitance is the sum of various capacitances shown in the second embodiment, that is, the drain / substrate junction capacitance Cj_db, part of the substrate capacitance, drain wiring / substrate coupling capacitance Cc_db, and drain coupling substrate capacitance Csub_db. Only the layout components are extracted from the overall capacity shown in FIG.
Further, the layout component Cgb_layout of the gate / substrate capacitance is obtained from the total capacitance indicated by the sum of the intrinsic gate / substrate capacitance Cgb_int, the gate wiring / substrate coupling capacitance Cc_gb, and the gate coupling substrate capacitance Csub_gb shown in the second embodiment. Only the layout components are extracted and shown.

Next, the estimation of the layout component will be specifically described by taking a high-frequency transistor having a multi-finger gate as an example.
FIG. 11A is a plan view of the first layout, FIG. 11B is a plan view of the second layout, and FIG. 11C is a chart showing the magnification of the parameter change accompanying the layout change. In these drawings, a layout change for doubling the gate finger length is shown.

First, a transistor with a multi-finger gate structure will be briefly described.
In the transistor of this example, a transistor active region 100 and a P-type impurity region (hereinafter referred to as a substrate contact region) 101 for substrate contact (back bias supply) are formed in a P-type semiconductor substrate (or P-well). ing. The active region 100 and the substrate contact region 101 have their geometric shapes (patterns) defined by forming an element isolation insulating layer 102 having a predetermined pattern on the surface portion of the semiconductor substrate.

  A gate electrode 103 (denoted by “G” in the drawing) having finger portions F1, F2,... Crossing the active region 100 is formed of, for example, polysilicon. Since the finger portions F1, F2,... Are orthogonal to the active region 100, the width of the active region 100 defines an effective gate length (a so-called gate width in a transistor) Wfinger of the gate finger portion. Hereinafter, the dimension Wfinger is unified with the name “finger length”. In the first layout of this example, the finger length Wfinger is 1 μm, and in the second layout, the finger length is 2 μm. The total effective gate width of the transistors needs to be the same before and after the layout change. Therefore, if the number of finger portions is 100 in the first layout, the number of finger portions is 50 in the second layout. Halved.

N-type impurities are introduced into the active region 100 by ion implantation using the gate electrode 103 as a mask, whereby N-type source regions S and drain regions D are alternately formed in the longitudinal direction of the active region 100.
A source wiring 104 made of a first metal is connected to the source region S, and a drain wiring 105 made of a first metal is connected to the drain region D. Although the source wiring 104 and the drain wiring 105 are drawn out in the same direction in the drawing, they may be drawn out in different directions alternately.

  On the other hand, the gate electrode 103 is provided with a predetermined number of gate contacts 106 for connecting it to the upper metal layer. In particular, when the source wiring 104 and the drain wiring 105 are drawn out in different directions, the gate electrode 103 is connected to a second metal gate wiring (not shown) through the gate contact 106.

  In FIG. 11A and FIG. 11B, layout components that can be estimated from the layout include a gate resistance layout component Rg_layout, a gate / source capacitance layout component Cgs_layout, a gate / drain capacitance layout component Cgd_layout, Layout component Csb_layout of substrate capacitance, layout component Cdb_layout of drain / substrate capacitance, layout component Cgb_layout of gate / substrate capacitance, layout components Rsub1_layout to Rsub4_layout of substrate resistances Rsub1 to Rsub4 (in the figure, Rsuby_layout (y = 1,2,3, 4)).

The meaning of these parameters has been described above and will not be repeated. Here, it is examined how these parameters change by doubling the finger length Wfinger.
Regarding the layout component Rg_layout of the gate resistance, if the finger length is doubled per finger, the polysilicon resistance is doubled and the number of contacts per unit resistance is reduced. Is done. Since the number of fingers is halved as a whole, it is necessary to further double the rate of change per finger. Therefore, the layout component Rg_layout of the gate resistance is [4 times + (extraction amount) × 2] before the change after the layout change.

  Regarding the layout component Rsuby_layout of the substrate resistance, it is considered that when the finger length is doubled, the average distance from the source region S or drain region D to the substrate contact region 101 is also roughly doubled. For this reason, the layout component Rsuby_layout of the substrate resistance is doubled per finger by the layout change, and is quadrupled as a whole.

As described above, the resistance doubles when the number of fingers is reduced by half, but in contrast, the capacity decreases by half when the number of fingers decreases by half.
Specifically, the layout component Cgs_layout of the gate / source capacitance, the layout component Cgd_layout of the gate / drain capacitance, the layout component Csb_layout of the source / substrate capacitance, and the layout component Cdb_layout of the drain / substrate capacitance are doubled per finger by changing the layout. , The total is 1 time. On the other hand, the layout component Cgb_layout of the gate / substrate capacitance becomes 1 time per finger and 1/2 times as a whole due to the layout change.
The parameter change magnifications described above are collectively shown in the chart of FIG.

In step ST13 shown in FIG. 10, a component having a layout dependency smaller than the layout dependency, such as an NQS component or an electromagnetic field effect component, is extracted from the first layout. To do. A component that does not have (almost) such layout dependency or is not important is hereinafter referred to as a “non-layout dependent component”.
FIG. 12 shows a specific method for extracting non-layout-dependent components using the gate resistance Rg and the substrate resistance Rsub1 as an example.
Here, it is assumed that the non-layout dependent component does not depend on the layout at all. However, when it is recognized that it is slightly dependent, it is possible to apply a correction coefficient of around “1” obtained empirically to the obtained non-layout dependent component.
As shown in FIG. 12A, when the finger length Wfinger is changed from 1 μm (first layout) to 2.5 μm and 5 μm (second layout), the layout component Rg_layout of the gate resistance is 1.09Ω. To 3.69Ω and 11.99Ω, respectively. In step ST11 described above, the total gate resistance Rg_total has already been obtained for the first layout. If the value is 11.40Ω, the non-layout dependent component, that is, the NQS component Rg_NQS and the electromagnetic field effect component Rg_em The sum can be calculated as 10.4Ω. Since this value is a constant value independent of the layout, it can also be applied to the second layout. Therefore, the total gate resistance Rg_total when the finger length Wfinger is 2.5 μm is calculated as 14.09Ω, and the total gate resistance Rg_total when the finger length Wfinger is 5 μm is calculated as 22.39Ω.

  As shown in FIG. 12B, when the finger length Wfinger is changed from 1 μm (first layout) to 2.5 μm and 5 μm (second layout), the layout component Rsub1_layout of the substrate resistance is 4 .10Ω to 25.63Ω and 102.50Ω respectively. In step ST11 described above, the total substrate resistance Rsub1_total has already been obtained for the first layout. If the value is 100.00Ω, the non-layout dependent component, that is, the NQS component Rsub1_NQS can be calculated as 95.9Ω. Since this value is a constant value independent of the layout, it can also be applied to the second layout. Therefore, the total substrate resistance Rsub1_total when the finger length Wfinger is 2.5 μm is calculated as 121.53Ω, and the total substrate resistance Rsub1_total when the finger length Wfinger is 5 μm is calculated as 198.40Ω.

In step ST14 shown in FIG. 10, the parameters extracted by the above-described method are set, and the breakdown and remaining parameters are determined.
Specifically, the intrinsic gate / substrate capacitance Cgb_int, the intrinsic gate / drain capacitance Cgd_int, and the intrinsic gate / source capacitance Cgs_int are set in the compact model, for example, BSIM3ver. A command value (op value) of 3 is used. Further, here, on the assumption that the layout pattern of the gate finger portion is changed and the layout of the other wiring is not changed, the inductance component Ld_layout of the drain wiring including the pad inductance and the inductance component Lg_layout of the gate wiring are constant values. For example, the pH is set to 34 pH, and the inductance component Ls_layout of the source wiring is set to a constant value, for example, 0.003 pH.
In addition, parameter values that are assumed to change due to a layout change are calculated or extracted again by a device simulator if necessary. For example, the slope in the Id-Vd characteristic is very important for analog circuits, but since this value should be changed by changing the finger length Wfinger, the parameter related to this slope depends on the finger length. To change.

  When all necessary parameters are set, in the next step ST15, whether or not the set parameter values are appropriate is determined by performing circuit operation (device simulation) using those values, or by using high frequency characteristics (S parameters, currents). Verification is made by measuring the gain cutoff frequency ft and the maximum operating frequency fmax). Thereby, the high frequency characteristic can be predicted.

FIGS. 13A to 13D are Smith charts showing a comparison between the S parameter calculated by the above method and measured data. In these figures, the measured value (actual value) of the S parameter of the actual device and the simulation result (“Sim.”) After setting the parameter in the transistor model of the present embodiment are shown. A high-frequency transistor having a multi-finger gate has a finger length Wfinger of each gate finger portion of 2.5 μm, a number of fingers of 40, a width (gate length) Lg of each gate finger portion of 0.07 μm, and a gate-drain capacitance ( Actual measurement value) is 0.025 pF. Further, the measurement frequency is changed from 100 MHz to 50 GHz.
From the figure, it can be seen that both agree well, and it is understood that the above method is appropriate. It can be confirmed that a scalable high-frequency transistor model related to the finger length Wfinger is constructed by taking the above method.

  According to the present embodiment, since the layout-dependent component is clear, once a high-frequency transistor model of a certain layout is created, when the layout is changed next, which parameter of the high-frequency transistor model is changed to what extent. It is clear what to do. As a result, the effort of model re-creation (change) associated with the layout change is minimized, and there are benefits that human and time costs can be reduced and hardware resources can be effectively used.

[Fourth Embodiment]
The present embodiment relates to a technique for optimizing the parameters of a high-frequency transistor.
In this optimization method, sensitivity analysis is performed to identify parameters that contribute to maintaining and improving high-frequency characteristics, and parameters to be optimized are identified from the results.

FIG. 14 is a table summarizing the sensitivity analysis results.
Here, sensitivity analysis was performed on 23 parameters shown in FIG. The parameters that have not been described so far include the capacitance Cgb_layout with the substrate of the gate extraction portion mainly composed of the gate electrode portion other than the gate finger portion, and the in-substrate resistance Rsub_gb_layout connected in series with this capacitance. Here, the capacitance Cgb_layout is calculated using a capacitance simulator “SENECA”, and the in-substrate resistance Rsub_gb_layout is calculated using a substrate simulator “substrate storm”. Since the other parameters have already been described, the description will not be repeated here.

For these 23 parameters, the finger length Wfinger was determined in three patterns of 1.0 μm, 2.5 μm and 5.0 μm. In the sensitivity analysis, for each of the current gain cutoff frequency ft and the maximum operating frequency fmax, the frequency difference is used as an index of how much frequency difference occurs when the finger length Wfinger is 1.0 μm and 5.0 μm. What was standardized by the center value was used.
As a result, the parameters that have a large effect on the current gain cutoff frequency ft due to the layout change are the capacitance Cgb_layout with the substrate of the gate extraction portion, the intrinsic gate / drain capacitance Cgd_int, and the gate wiring / substrate coupling capacitance Cc_gb in descending order of the influence. I found out that It was also found that the parameters having a large influence on the maximum operating frequency fmax are the total gate resistance Rg_total, the intrinsic gate / drain capacitance Cgd_int, and the substrate resistance Rsub1 in the order of the large influence.
In addition to the above, the S-parameter may be used as the high-frequency characteristic for performing the parameter sensitivity analysis.

  According to the present embodiment, how to improve the layout based on the sensitivity analysis result of the parameters with respect to the high frequency characteristics, based on the main parameters that affect the high frequency characteristics, such as the current gain cutoff frequency ft and the maximum operating frequency fmax. You can easily see if it is good. As a result, layout improvement points can be predicted in advance, and the labor involved in the layout change can be minimized, so that human and time costs can be reduced and hardware resources can be effectively used.

  In the above first to fourth embodiments, it has been mentioned that some parameters have frequency dependency. In addition, some key parameters can be biased. The method will be described below.

In general, in the existing compact model, since the capacitance component (capacitance parameter) is not appropriate, an error occurs when operated with any bias (bias voltage or bias current). Here, the error component is obtained for each bias.
Specifically, the high-frequency transistor model in which the parameters extracted in the above embodiment are set is operated with a certain bias using, for example, a device simulator, and the parameter value obtained at that time and the parameter value of the compact model are And find the difference. This operation is performed with all biases (more practically, discrete representative points) for all or necessary parameters. The difference or the correct parameter value is stored for each bias, for example, as a table. When operating an actual high-frequency transistor model with a simulator, etc., depending on the required bias during operation, the difference between the parameter values corresponding to the bias or the correct parameter value is read, and the value of each parameter is automatically Correct with. In addition, when using a difference, this difference is added to the existing parameter value of a compact model. On the other hand, when the correct parameter value is used, the existing parameter value of the compact model is set to zero and is effectively replaced with the correct parameter value.

  Typical parameters that cause an error if such bias dependence is not given include the intrinsic gate / drain capacitance Cgd_int and the intrinsic gate / source capacitance Cgs_int. If necessary, other parameters can be biased.

  In this way, by making the necessary parameters bias dependent, the high-frequency transistor model incorporated in the simulator using the so-called compact model (BSIM3 ver. 3, MOS model 9, EKV) that operates with any bias can be operated more appropriately. Simulation with high accuracy. Here, as long as the high-frequency transistor model is used, the type of simulator and the type of compact model are arbitrary and not essential.

Further, although power consumption is not mentioned in the first to fourth embodiments described above, particularly in a CR series circuit composed of a substrate coupling capacitance and a substrate resistance, the power consumption is further increased from the peak point of power consumption. It is desirable to set these parameters to operate in a low region.
More specifically, a CR series circuit is constituted by the gate wiring / substrate coupling capacitance Cc_gb and the gate coupling substrate resistance Rsub_gb shown in FIG. 1, and a CR series is constituted by the drain wiring / substrate coupling capacitance Cc_db and the gate coupling substrate resistance Rsub_db. A circuit is configured. In these CR series circuits, the power consumption has a peak, and the substrate resistance at the peak can be calculated by 1 / (2πC). Here, “C” is the value of the gate wiring / substrate coupling capacitance Cc_gb or the drain wiring / substrate coupling capacitance Cc_db. In order to suppress power consumption in the substrate in this embodiment mode, it is desirable that power consumption be lower than the peak point. In addition, there are restrictions on the wiring structure and substrate resistance, etc., but within this restriction range, the gate wiring and wiring can be reduced so that the substrate resistance is as far as possible from the peak point, or the power consumption is sufficiently reduced. It is desirable to set the value of the substrate coupling capacitance Cc_gb or the drain wiring / substrate coupling capacitance Cc_db.

  In the present embodiment, prediction of characteristics can be easily achieved in addition to the benefit of increasing the accuracy of simulation as described above. In other words, since it was not possible to know the effects of transistor model parameters on the characteristics in the past, we made a prototype (actual device fabrication), measured the characteristics, and fitted the measurement results to the desired characteristics. To find the correct parameters. On the other hand, in this embodiment, if this parameter is optimized, the influence of each parameter on the high-frequency characteristics can be known to some extent. The benefit is that the characteristics can be predicted. In addition, the cause for optimizing the required characteristics is clarified, and the layout can be easily optimized.

  The present invention is used by being incorporated in software such as simulation of a high-frequency circuit and device simulation of a high-frequency transistor unit, and can be applied to the use of a high-frequency transistor model for expressing a transistor unit on a computer.

It is an equivalent circuit diagram of the high frequency transistor model according to the embodiment of the present invention. It is explanatory drawing of the NQS effect which shows the coupling | bonding and channel resistance of the gate electrode and channel in a high frequency as a distribution constant. FIG. 5 is an equivalent circuit diagram showing a schematic configuration when an intrinsic transistor portion is an equivalent circuit model. (A)-(D) are the graphs of the S parameter which show the simulation result using the high frequency transistor model which concerns on 1st Embodiment in comparison with an actual value. It is a graph which shows the frequency characteristic of the gate resistance value extracted from an actual S parameter. It is a circuit diagram which extracts and shows a parameter required in order to suppress the influence of the increase in the output resistance in the high frequency by the characteristic adjustment by the conventional fitting. It is a Smith chart which shows the locus | trajectory of the output reflection coefficient in the circuit shown in FIG. It is a table | surface which compares and shows the board | substrate resistance value (measured value) calculated | required by calculation from the layout, and the board | substrate resistance value extracted after fitting of S parameter. It is a flowchart which shows the main steps for determining a parameter value with the preparation method of the high frequency transistor model concerning a 2nd embodiment. It is a flowchart which shows the outline of the re-creation (update) method of the model accompanying the layout change in the high-frequency transistor model creation method according to the third embodiment. (A) is a plan view of the first layout, (B) is a plan view of the second layout, and (C) is a chart showing the magnification of the parameter change accompanying the layout change. It is explanatory drawing which shows the specific value of a non-layout dependence component by making a gate resistance and a substrate resistance into an example. (A)-(D) are Smith charts which compare the S parameter calculated by the method which concerns on 3rd Embodiment, and measured data. It is the table | surface which put together the sensitivity analysis result by the preparation method of the high frequency transistor model which concerns on 4th Embodiment. It is sectional drawing of the high frequency transistor described in the nonpatent literature 1. FIG. 16 is an equivalent circuit diagram showing an equivalent circuit model including an external circuit that reproduces the operation of the high-frequency transistor shown in FIG. 15.

Explanation of symbols

  Csub_db: drain coupling substrate capacitance, Csub_gb ... gate coupling substrate capacitance, Cc_db ... drain wiring / substrate coupling capacitance, Cc_gb ... gate wiring / substrate coupling capacitance, Cc_gd ... gate / drain wiring coupling capacitance, Cc_gs ... gate / source wiring coupling capacitance, Cgb_int: intrinsic gate-substrate capacitance, Cgd_int: intrinsic gate-drain capacitance, Cgs_int: intrinsic gate-source capacitance, Cgd_ov: gate-drain overlap coupling capacitance, Cgs_ov: gate-source overlap coupling capacitance, Cj_db ... drain-substrate junction capacitance, Cj_sb ... source-substrate junction capacitance, D0 ... external drain terminal, G0 ... external gate terminal, Ld_layout ... drain wiring inductance component, Lg_layout ... gate wiring inductance component, Ls_layout ... source wiring inductance component , Mi ... intrinsic transistor part, Rsub1 ... Drain substrate resistance, Rsub2, Rsub3 ... Source / drain substrate resistance, Rsub4 ... Source substrate resistance, Rsub1y_layout (y = 1,2,3,4) ... Layout component of substrate resistance, Rsuby_NQS (y = 1,2,3 4) ... NQS component of substrate resistance, Rsub_db ... drain coupling substrate resistance, Rsub_gb ... gate coupling substrate resistance, Rch ... channel resistance, Rd ... drain resistance, Rd_layout ... layout component of drain wiring resistance, Rds ... output resistance, Rg_NQS ... NQS gate resistance, Rg_em: electromagnetic field effect component of gate resistance, Rg_layout: layout component of gate resistance, Rg_total: total gate resistance, Rg0: gate resistance, Rs: source resistance, Rs_layout: layout component of source wiring resistance, S0: external Source terminal, bi... Intrinsic back bias node, di... Intrinsic drain node, gds... Drain conductance, gi. Tonodo, gm ... transconductance, si ... intrinsic source node

Claims (2)

When creating a transistor model, which is circuit data including changeable parameters, for a transistor unit including an intrinsic transistor part and electrodes, contacts, and wirings connected to the intrinsic transistor part and changing according to the layout. The circuit parameters included in the device model of the intrinsic transistor unit are obtained from a device simulation by a device simulation apparatus using the device model and a DC measurement by a DC measurement apparatus, thereby obtaining an intrinsic parameter of the intrinsic transistor unit. A first step of determining parameters;
A second step of obtaining a non-quasi-static parameter indicating a time delay of a carrier traveling in the channel of the intrinsic transistor part and a resistance component depending on a skin effect by a device simulation using a device simulation device ,
Depending on the layout, the parasitic parameters of the electrodes from the intrinsic transistor part to the source external terminal, the drain external terminal, and the gate external terminal of the transistor unit, or the wiring connected to the electrodes via contacts A changing resistance value and an inductance value are obtained from device simulation by a device simulation apparatus, and at this time, with respect to gate resistance, a third step using a combined resistance of the non-quasi-static parameter and the resistance component depending on the skin effect;
A fourth step of obtaining a capacitance value independent of layout by device simulation using a device simulation apparatus;
including,
An intrinsic parameter of the intrinsic circuit, an extrinsic parameter including a resistance component dependent on the non-quasi-static parameter and the skin effect, and a layout dependent resistance and parasitic parameter of the inductor,
How to create a high-frequency transistor model. When at least one of the non-quasi-static parameter and the resistance component depending on the skin effect includes a frequency-dependent component that changes at a non-negligible rate as the operating frequency increases, in the device simulation of the third step , device simulation device, wherein at least one of the non-quasi-static parameters and the resistance component dependent on the skin effect, the method of creating a high-frequency transistor model according to claim 1 that gives a function representing the frequency dependent.

Images