Device Simulation Advances in 1993

Accelerated development of new device simulation capabilities occurred during 1993. Numerous features were added to the existing foundation provided by ATLAS for 2D simulation and by THUNDER for 3D simulation. In this article we summarize some of the major developments that took place in 1993, and we preview some of the capabilities that will become available in 1994.


Blaze was introduced to meet the demand for a full-featured simulator of devices fabricated using advanced and novel semiconductor systems. Blaze handles general semiconductors and graded or abrupt heterojunctions. It provides advanced non-local energy balance models of charge transport, as well as the local drift-diffusion model. Working in combination with other ATLAS products, Blaze accounts for the effects of lattice heating, heatsinks, optoelectronic interactions, and polycrystalline and amorphous materials. Blaze is supplied with a library of material parameters for more than 40 semiconductors, and users can add additional materials.The Blaze development group includes specialists in III-V, heterojunction, and optoelectronic device physics. Development also benefits from collaborations with leading universities such as the University of Toronto and the University of Virginia.

Blaze is fully integrated with the Flash-2D GaAs process simulator, and with the interactive tools that support Silvaco's core simulators. Flash-2D and Blaze can be used within Virtual Wafer Fab.

Figure 1 depicts a pseudomorphic HEMT structure based on a GaAs-AlGaAs-InGaAs material system. Electron concentration contours are also presented on this plot for almost fully depleted conditions with a gate voltage of Vg=-1.2V, and a drain voltage Vd=4V. The family of drain current versus drain voltage characteristics for different gate voltages for this structure are shown in Figure 2 .


Figure 1. GaAs-AlGaAs-InGaAs pseudomorphic HEMT structure and electron contours for bias conditions of Vg=-1.2V, Vd=4V. Figure 2. GaAs-AlGaAs-InGaAs HEMT Id - Vd characteristics for different gate voltages.


Blaze is now the unchallenged leader for simulating devices fabricated using advanced and novel material systems. No competing simulator provides as many capabilities, is being developed as aggressively, is supported as expertly, is interfaced to a GaAs process simulator, or works within Virtual Wafer Fab!



The application of simulators to advanced material systems is often challenging because these systems are less well understood and characterized than silicon. Researchers need the flexibility to develop and test new physical models, and to define structures that have additional degrees of freedom, such as position-dependent composition. Providing users with source code is not viable, due to issues such as security, support, and version proliferation. Silvaco has overcome these problems by developing C-Interpreter technology that allows users to define models as C source code that is interpreted at run time.



Another major milestone achieved during 1993 was the development and release of the first commercial simulator of semiconductor lasers. Silvaco had previously established leadership in optoelectronic device simulation with Luminous, which works as part of ATLAS to simulate non-coherent optoelectronic interactions, and includes automatic structure-resolving ray tracing. Laser provides the additional capabilities needed to simulate semiconductor lasers.

Laser self-consistently solves the Helmholtz equation, to obtain optical fields and photon densities, and calculates carrier recombination due to light emission (i.e. stimulated emission). Two models for optical gain are provided, and single-frequency and multi-mode analysis can be performed. Figure 3 shows spectral characteristics for a ridge waveguide laser diode for three different bias conditions, and Figure 4 shows the light intensity distributions for a bias voltage of 1.6V.

Figure 3. Spectral characteristics of the "ridge" waveguide laser diode under different bias conditions. Figure 4. Light intensity distribution in the "ridge" waveguide laser diode with a bias voltage of 1.6V.


Reliability Model

Physically-based reliability simulation models were implemented in ATLAS. These models account for the injection of hot electrons and holes into the gate oxide, and trapping on defect sites in the oxide, under DC or dynamic stress conditions. Figures 5 and 6 illustrate results obtained for an LDD NMOS transistor with a gate oxide thickness of 200, stressed with Vd= Vg= 5V for 1000s. Figure 5 shows the density of trapped electrons plotted as a function of position along the interface at different times. Figure 6 shows the the pre- and post-stress gate characteristics of the transistor. The effect of stressing on circuit performance can be established using stressed numerical devices in MixedMode.


Figure 5. Evolution of the density of trapped electrons. Figure 5. Evolution of the density of trapped electrons.


Dynamic Traps

The ability to simulate dynamic traps was added to ATLAS. Deep level traps strongly influence the performance of GaAs MESFET's and other devices fabricated using modern material systems and polycrystalline and amorphous materials. The effects associated with deep level traps include transconductance reduction, drain I-V hysteresis, modified breakdown behavior, modified transient switching characteristics, low-frequency dispersion, and backgating. Figure 7 shows an example of drain and gate terminal current transients calculated for a GaAs MESFET with a 1m gate length, fabricated on a semi-insulating Cr-doped substrate. The long time constant changes are due to the charging and discharging of deep level traps.

Figure 7. Transient characteristics of gate MESFET
fabrication on semi-insulted substrate.


MixedMode Enhancements

The MixedMode circuit simulator can use numerical ATLAS devices as well as compact circuit models. The first version of MixedMode was introduced at the end of 1992. The capabilities of MixedMode were enhanced greatly during 1993. Many additional electrical models were introduced, and user-defined models are now supported through the C-Interpreter. An interface to Luminous provides unique capabilities for simulating electrically and optically modulated optoelectronic devices in real circuit environments. In addition, MixedMode now provides full support for AC analysis.


Improved Transient Calculations

Several algorithmic improvements that speed up transient analysis were developed. An option for obtaining an initial guess at the next time level by projection is now available. This provides speed-ups of 5-7 for some types of problem. Improved error control for quasi-static situations was added, and this provides orders of magnitude faster calculation of certain situations, such as programming and erasing of EPROM devices.


Fast Mode Calculations

For idealized (planar) geometries and pre-defined operating modes (bipolar or MOS), it is practical to automate gridding, bias stepping, and parameter extraction. ATLAS and DevEdit now support FastMOS and FastBIP modes of operation that provide this automation. These modes are especially useful in the Virtual Wafer Fab. The FastMOS mode also provides replacement capabilities for S-MINIMOS.


Enhanced 3-D Capabilities

THUNDER was made more versatile during 1993. The existing version of THUNDER was made more modular by partitioning it into a framework for structure specification and results visualization, and a device simulation module. New modules were added to support calculation of the properties of 3D interconnects, and the thermal properties of 3D structures. The three modules were named (rather unimaginatively) Device3D, Interconnect3D, and Thermal3D, respectively.

Interconnect3D calculates the capacitances and resistances associated with general interconnect structures. It replaces EXTRACT, which is no longer offered by Silvaco. Interconnect3D retains the highly accurate energy method for calculating capacitances. Compared to other interconnect calculation programs it provides several advantages: it handles realistic non-planar, non-rectangular structures; and it uses a prismatic mesh that typically requires many fewer grid points than the tensor-product meshes used by older programs. Interconnect3D is proving particularly popular with groups that perform both 3D device simulation and 3D parasitic extraction, since a common interface is used for both types of calculation.

THERMAL calculates the temperature distributions associated with 3-D thermal environments, including heat sources. Very general thermal boundary conditions are supported. As a result of inputs from customers, the capabilities of THERMAL are being extended to include time-dependent behavior, time-dependent heat sources, and temperature dependent thermal conductivities. Additional convenience features will support the simulation needs of IC designers, power transistor designers, and packaging specialists.


'Next Generation' Capabilities

A version of ATLAS that can account simultaneously for non-local and non-isothermal effects will be released during 1994. This product completes a natural progression of escalating functionality. Prior to the 1980's virtually all device simulation used the 'local' drift-diffusion transport model, and assumed an isothermal lattice. During the 1980's non-local models of charge transport (e.g., energy balance models) were developed, but these retained the isothermal lattice approximation. During the early 1990's the isothermal lattice assumption was removed in codes that retain the drift-diffusion approximation. It was natural that attention would turn to removing both the drift-diffusion and the isothermal approximations.

A device simulator that accounts for both non-local and non-isothermal effects can be a workhorse simulator until well into the twenty-first century. A project to develop such a simulator was therefore initiated by Silvaco. It was decided that this next generation simulator would be versatile with respect to numerical techniques as well as physical models. Expert users should, for any set of physical approximations, have access to a full Newton scheme and built-in and user-specifiable block iteration schemes.

A completely new six-equation core solver for ATLAS was defined and implemented to these specifications. Modern software techniques were exploited to achieve benefits such as dynamic memory allocation. Implementation of the new six equation solver is complete, and the code is being subjected to exhaustive testing.

Some results calculated for bipolar and SOI devices are presented in Figures 8 and 9. These results show clearly the significant differences that can be obtained between results calculated using an isothermal drift-diffusion model (DD); a non-isothermal drift-diffusion model (NDD); an isothermal energy balance model (EB); and a non-isothermal energy balance model (NEB).


Figure 8. BJT breakdown characteristics calculated using different models. Figure 9. SOI breakdown characteristics calculated using four different models.


The new version of ATLAS will be released, with completely revised and updated documentation during 1994. Users of Silvaco software will then be using ATHENA, the most advanced process simulator, to provide input to ATLAS, the most advanced device simulator; and both products will be used within Virtual Wafer Fab. Everything needed for next generation simulation will be in place.