This paper reports the application results of the state-of-the-art advanced process/device modeling to the 0.13 [µm] CMOS design solution. It has been demonstrated that the S/D-extension junction depth, the well profile, the channel profile and the drive current of the 0.13 [µm] CMOS can be predicted with reasonable accuracy. Further model improvement is required to predict the ΔL and the Vt-Lg characteristics of the devices with the tilted pocket I/I more accurately. It is quite beneficial to construct several design maps by using the state-of-the-art advanced TCAD in a 'carpet bombing' way in the early stage of the development of new generation CMOS.

A systematic experimental and modeling study is reported, which characterizes the low-frequency noise spectrum of 100 nm-MOSFETs accurately. Two kinds of measured spectra are observed: 1/f and non-1/f spectra. The non-1/f spectrum is analysed by forward and backward measurements with exchanged source and drain, and shown to be due to a randomly distributed inhomogeneity of the trap density along the channel and within the gate oxide. By averaging the spectra of identical MOSFETs on a wafer the measured non-1/f noise spectra reduce to a 1/f characteristics. On the basis of these measurement data a noise model for circuit simulation is developed, which reproduces the low-frequency noise spectrum with a single model parameter for all gate lengths and under any bias conditions.

Small-size MOSFETs are becoming core devices in RF applications because of improved high frequency characteristics. For reliable design of RF integrated circuits operating at the GHz range, accurate modeling of small-size MOSFET characteristics is indispensable. In MOSFETs with reduced gate length (Lg), the lateral field along the MOSFET channel is becoming more pronounced, causing short-channel effects. These effects should be included in the device modeling used for circuit simulation. In this work, we investigated the effects of the field gradient in the gate-drain capacitance (Cgd). 2-Dimensional (2D) simulations done with MEDICI show that the field gradient, as it influences the channel condition, induces a capacitance which is visible in the MOSFET saturation operation. Changes in Cgd is incorporated in the modeling by an induced capacitance approach. The new approach has been successfully implemented in the surface-potential based model HiSIM (Hiroshima-university STARC IGFET Model) and is capable of reproducing accurately the measured Cgd-Lg characteristics, which are particularly significant for pocket-implant technology. Results show that pocket-implantation introduces a steep potential increase near the drain region, which results to a shift of the Cgd transition region (from linear to saturation) to lower bias voltages. Cgd at saturation decreases with Lg due to steeper surface potential and increased impurity concentration effects at reduced Lg.

We have developed a model for circuit-simulation which describes the MOSFET region from pinch-off to drain contact based on the surface potential. The model relates the surface-potential increase beyond the pinch-off point to the channel/drain junction profile by applying the Gauss law with the assumption that the lateral field is greater than the vertical one. Explicit equations for the lateral field and the pinch-off length are obtained, which take the potential increase in the drain overlap region into account. The model, as implemented into a circuit simulator, correctly reproduces measured channel conductance and overlap capacitance for 100 nm pocket-implant technologies as a function of bias condition and gate length.

The urgent tasks of MOSFET modeling for circuit simulation are easy adaptation to new physical phenomena arising for advancing technologies, and, of course, sufficient simulation accuracy. Approaches currently being pursued for developing such MOSFET models are summarized. Their capabilities for accomplishing these tasks as well as the important remaining problems are discussed. Main focus is given on the model HiSIM, the first commonly available model based on the drift-diffusion approximation developed for 0.10 µm MOSFET technology node.

Numerical simulation of multiple semiconductor devices is necessary to analyze dynamic two- and three-dimensional interactions among devices in CMOS inverters, SRAM cells, and other more complicated gates. With the advent of complete 3D process simulation, an alternative to brute-force transient device simulation must be found for large contiguous silicon regions. Our approach differs from brute-force methods in that we focus not on the time-step control but rather on the latency in the system. By latency, we do not mean that activity within parts of the simulation has ceased or has reached a steady state. Rather, we imply that a simpler form of the solution, a macromodel, can be used to decouple the problem into smaller subproblems. This means that while integrating at a particular time-step, the system of device equations needs only to be solved for a subset of nodes, whereas the node device variables approximated by macromodels are treated as fixed boundary conditions. This drastically reduces the size of the system of equations to be solved at each time-step and allows each node to have a different time-step. Because the responses have exponential-like behavior, we aim to approximate carrier and potential values with closed-form exponential macromodels during a time interval. To assure the accuracy of the simulation, we implement several error formula which predict the range of validity of this interval. Moreover, this approach takes advantage of a standard workstation environment (e.g. SparcStation, DECstation, RS6000). This method has been successfully exploited in circuit simulators like SAMSON, which relies on a sophisticated predictor/corrector scheme based on Gear's backward-differentiation formulae (BDF) and depends on partitioning the circuit by inspection. The device simulation problem differs because the partitioning can not be performed by inspection, and the overhead of implementing multi-order BDF would negate the advantage of the decoupling. Instead, we propose the event-driven simulator, AWETOPSY (Asymptotic Waveform Evaluation for Transient Optimized and Partitioned Simulation) that uses automatic partitioning (domain decomposition) and a straightforward second-order integration scheme that we call the power method in conjunction with exponentially-based macromodeling of Asymptotic Waveform Evaluation to exploit the latency. Although Asymptotic Waveform Evaluation (AWE) was originally developed to simplify the solution of linear circuits, we have adapted it to transient device simulation.

A compact model is proposed for accurately incorporating effects of STI (shallow trench isolation) stress into post-layout simulation by making layout-dependent corrections to SPICE model parameters. The model takes in-plane (longitudinal and transverse) and normal components of the layout-dependent stress into account, and model formulas are devised from physical considerations. Not only can the model handle the shape of the active-area of any MOSFET conforming to design rules, but also considers distances to neighboring active-areas. Extraction of geometrical parameters from the layout can be performed by standard LVS (layout versus schematic) tools, and the corrections can subsequently be back-annotated into the netlist. The paper spells out the complete formulation by presenting expressions for the mobility and the threshold voltage explicitly by way of example. The model is amply validated by comparisons with experimental data from 90 nm- and 65 nm-CMOS technologies having the channel orientations of, respectively, <110> and <100>, both on a (100) surface. The worst-case standard errors turn out to be as small as 1.7% for the saturation current and 8 mV for the threshold voltage, as opposed to 20% and 50 mV without the model. Since device characteristics variations due to STI stress constitute a significant part of what have conventionally been treated as random variations, use of the proposed model could enable one to greatly narrow the guardbands required to guarantee a desired yield, thereby facilitating design closure.

In-place AC measurements of the signal gain and substrate sensitivity of differential pair transistors of an analog amplifier are combined with DC characterization of the threshold voltage (Vth) of the same transistors. An on-chip continuous time waveform monitoring technique enables in-place matrix measurements of differential pair transistors with a variety of channel sizes and geometry, allowing the wide coverage of experiments about the transistor-level physical layout dependency of substrate noise response. A prototype test structure uses a 90-nm CMOS technology and demonstrates the geometry-dependent variation of substrate sensitivity of transistors in operation.

Layout-aware compact models proposed so far have been generally verified only for simple test patterns. However, real designs use much more complicated layout patterns. Therefore, models must be verified for such patterns to establish their practicality. This paper proposes a methodology and test patterns for exhaustively and systematically validating layout-aware compact models for general layout patterns for the first time. The methodology and test patterns are concretely shown through validation of a shallow trench isolation (STI) stress compact model proposed in [1]. First, the model parameters for a 55-nm CMOS technology are extracted, and then the model is verified and established to be accurate for the basic patterns used for parameter extraction. Next, fundamental ideas of model operation for general layout patterns are verified using various verification patterns. These tests revealed that the model is relatively weak in some cases not included in the basic patterns. Finally, the errors for these cases are eliminated by enhancing the algorithm. Consequently, the model is confirmed to have high generality. This methodology will be effective for validating other layout-aware compact models for general layout patterns.

A continuous-time waveform monitoring technique for quality on-chip power noise measurements features matched probing performance among a variety of voltage domains of interest in a VLSI circuit, covering digital Vdd, analog Vdd, as well as at Vss, and multiple probing capability at various locations on power planes. A calibration flow eliminates the offset as well as gain errors among probing channels. The consistency of waveforms acquired by the proposed continuous-time monitoring and sampled-time precise digitization techniques is ensured. A 90-nm CMOS on-chip monitor prototype demonstrates dynamic power supply noise measurements with 200 mV at 2.5 V, 1.0 V, and 0.0 V, respectively, with less than 4 mV deviation among 240 probing channels.

This paper reports the evaluation results of the channel boron distribution in the deep sub-0.1 [µm] n-MOSFETs for the first time. It has been found that the boron depletion effect becomes dominant and the reverse short channel effect becomes less significant in the deep sub-0.1 [µm] n-MOSFETs. It has been also found that the sheet charge distribution responsible for the reverse short channel effect is localized within a distance of 100 [nm] from the source/drain-extension junction.

A single tone pseudo-noise generator with a harmonic-eliminated waveform is proposed for measuring noise tolerance of analog IPs. In the waveform, the harmonics up to the thirteenth are eliminated by combining seven rectangular waves with 22.5-degree spacing phases. The proposed waveform includes only high region frequency harmonic components, which are easily suppressed by a low-order filter. This characteristic enables simple circuit implementation for a sine wave generator. In the circuit, the harmonic eliminated waveform generator is combined with a current controlled oscillator and a frequency adjustment circuit. The single tone pseudo-noise generator can generate power line noise from 20 MHz to 220 MHz with 1 MHz steps. The SFDR of 40 dB is obtained at the noise frequency of 100 MHz. The circuit enables the measurement of frequency response characteristics measurements such as PSRR.

The response of differential pairs against low-frequency substrate voltage variation is captured in a combined transistor and substrate network models. The model generation is regularized for variation of transistor geometries including channel sizes, fingering and folding, and the placements of guard bands. The expansion of the models for full-chip substrate noise analysis is also discussed. The substrate sensitivity of differential pairs is evaluated through on-chip substrate coupling measurements in a 90 nm CMOS technology with more than 64 different geometries and operating conditions. The trends and strengths of substrate sensitivity are shown to be well consistent between simulation and measurements.