FPGA-SPICE: A Simulation-Based Architecture Evaluation Framework for FPGAs

In this paper, we developed a simulation-based architecture evaluation framework for field-programmable gate arrays (FPGAs), called FPGA-SPICE, which enables automatic layout-level estimation and electrical simulations of FPGA architectures. FPGA-SPICE can automatically generate Verilog and SPICE netlists based on realistic FPGA configurations and a high-level eTtensible Markup Language-based FPGA architectural description language. The outputted Verilog netlists can be used to generate layouts of full FPGA fabrics through a semicustom design flow. SPICE simulation decks can be generated at three levels of complexity, namely, full-chip-level, grid-level, and component-level, providing different tradeoff between accuracy and simulation time. In order to enable such level of analysis, we presented two SPICE netlist partitioning techniques: loads extraction and parasitic net activity estimation. Electrical simulations showed that averaged over the selected benchmarks, the grid-/component-level approach can achieve 6.1x/7.5x execution speed-up with 9.9%/8.3% accuracy loss, respectively, compared to the full-chip level simulation. FPGA-SPICE was showcased through three different case studies: 1) an area breakdown analysis for static random access memory-based FPGAs, showing that configuration memories are a dominant factor; 2) a power breakdown comparison to analytical models, analyzing the source of accuracy loss; and 3) a robustness evaluation against process corners, studying their impact on energy consumption of full FPGA fabrics.

A modular software package for the analysis of power networks and electrical drives

Alain Sapin, Jean-Jacques Simond

This paper deals with a software package for the numerical analysis in transient and steady-state modes of power electrical networks or variable speed drives with arbitrary topologies. The package is composed of a series of units, each representing a specific cell in the network : voltage supply, electrical machine, mechanical system, transmission line, circuit-breaker, phase shifting transformer, static converter with control and command organ, regulator, etc. SIMSEN is highly flexible and efficient. It is implanted on microcomputer or on workstation. The network or the variable speed drive to be simulated is assembled using a graphic input interface by adequatly choosing and linking the building units, so as to fulfill a desired topology. An existing system may be easily extended or modified. A simple procedure can be used in order to define a new unit or to modify an existing one. The initial conditions of operation can be partly or entirely specified by the user. A transient mode of operation may include several successive perturbations. The simulation results are displayed through an efficient graphic interface. An original feature of SIMSEN is its ability to analyse electrical networks involving semi-conductors (diodes rectifiers, thyristor or GTO current converters, voltage inverters, etc.). Thus, systems having a complex structure can be simulated. Practical examples of applications to such systems constitute the main part of this paper : constant or variable speed groups involving induction and synchronous machines, HVDC networks, reactive power static converter SVC. These examples show how SIMSEN can be used for an optimized design of complex networks.

1995

Numerical modeling and neural networks to identify constitutive parameters from in situ tests

Rafal Filip Obrzud

This study proposes a new advanced algorithm for determining material parameters based on in situ tests. In situ testing gives an opportunity to perform soil characterization in natural stress conditions on a representative soil mass. Most field techniques reduce soil disturbances to minimum, allowing investigating the response of virgin soil. Self-boring pressuremeter tests (SBPT) and standard piezocone tests (CPTU) are widely used to deduce properties of clayey soils through analytical and empirical correlations between soil properties and experimental measurements. Empirical correlations usually require some tuning based on reference laboratory data because first-order estimates for typical correlation coefficients may give unreliable evaluation of soil properties. Analytical correlations are mostly based on cavity expansion methods which are restricted to either fully drained or perfectly undrained problems, so that inverse closed-form solutions for relatively simple constitutive models can be derived. In practice, however, depending on physical and consolidation properties of the soil, partially drained conditions may occur during field testing, leading to an erroneous estimation of clay characteristics. Therefore, elaborating a generic parameter identification framework, which is based on the artificial neural network (NN) technique and which may improve the reliability of soil properties derived from in situ testing, is the main goal of this research. This study explores the possibility of using NNs to solve complex inverse problems including partially drained conditions. In other words, NNs are used to map experimental measurements onto set of soil properties. The development of NN-based inverse models is based on a training data sets which consists of pseudo-experimental measurements derived from numerical simulations of both the SBPT and the CPTU test in normally- and lightly overconsolidated clay type material. The study presents a generic two-level procedure designed for the calibration of constitutive models of soils. It is demonstrated that NN inverse models can be easily integrated into the classical back-analysis. At the first level, the NN approach is applied to achieve the first approximation of parameters. This technique is used to avoid potential pitfalls related to the conventional gradient-based optimization (GBO) technique, considered here as a corrector that improves predicted parameters. Trained NNs as parallel operating systems can provide output variables instantly and without a costly GBO iterative scheme. The proposed framework is verified for the elasto-plastic Modified Cam Clay (MCC) model that can be calibrated based on standard triaxial laboratory tests, i.e. the isotropic consolidation test and the consolidated isotropic drained compression test. The study presents formulations of the input data for the NN predictors, enhanced by a dimensional reduction of experimental data using principal component analysis (PCA). The determination of model characteristics is demonstrated, first on numerical pseudo-experiments and then on the experimental data. Furthermore, the efficiency of the proposed approach in terms of accuracy and computational effort is also discussed. The verified two-level strategy is applied to a numerical procedure of parameter identification for the boundary value problem (BVP) of the SBPT. The coupled hydro-mechanical finite element (FE) formulation allows the generated excess pore water pressure to be dissipated during simulations of the expansion test, followed then by a holding test. Numerical simulations demonstrate that volume changes that may occur in clay during the expansion test due to partial drainage, can cause local soil hardening near the cavity wall and affect parameter interpretation for pressuremeter tests. Therefore, the NN technique is applied to obtain an initial guess for model parameters, taking into account the possible partially drained conditions during the expansion test. Parameter identification based on measurements obtained through the pressuremeter expansion test and two types of holding tests is illustrated on the MCC model. NNs are trained using a set of synthetic test samples, which are generated by means of FE simulations based on constrained random permutations of input model parameters. The measurements obtained through expansion and consolidation tests are normalized by the proposed normalizing formulas so that NN predictors operate independently of the testing depth. Examples of parameter determination are demonstrated on both numerical data and field measurements from the Fucino clay deposit. The efficiency of the combined parameter identification in terms of accuracy, effectiveness and computational effort is also discussed. Finally, an application of NN predictors as a stand-alone support for soil profiling is presented for the piezocone test. By similarity to the SBPT problem, a number of NN inverse models are developed based on the results derived from rigorous FE analyzes. The FE model of piezocone penetration involves numerical formulation for the two-phase material obeying the MCC law and including the large strain theory, as well as the large deformation formulation for contact interface. It is demonstrated that a considerable computational effort related to the generation of the training database can be reduced by optimizing the mesh size and "steady-state" depth in function of soil rigidity index. Due to a severe loss of measurement accuracy observed in the finite elements adhered to the "rough" interface, an equivalent semi-numerical approach is proposed to account for frictional effects in different drainage conditions which are delineated from a number of numerical simulations. The validity of the developed penetration model is verified in detail by means of comparisons with other theoretical solutions and parametric studies synthesized from literature, as well as experimental evidence for both undrained and partially drained scenarios. An extended parametric study including influence analyzes of strength and stress anisotropy, rigidity index and cone roughness on two cone factors provides new insight into the analysis of cone penetration. The shortcomings of the FE model due to the limitations of the applied constitutive model are also discussed. As regards NN models, different configurations of input variables, including standard normalized piezocone metrics and other available soil characteristics are investigated in terms of feasibility of effective NN training. The post-training regression analyzes are performed for numerical data allowing the assessment of the influence of specific input variables on accuracy of parameters predictions. Finally, the developed NN models are applied to predict parameters based on field measurements for a number of characterization sites. Provided examples demonstrate that NN inverse models may constitute an effective complementary support during the first-order quantification of the MCC parameters from piezocone measurements.

EPFL2009

Circuit Design, Architecture and CAD for RRAM-based FPGAs

Xifan Tang

Field Programmable Gate Arrays (FPGAs) have been indispensable components of embedded systems and datacenter infrastructures. However, energy efficiency of FPGAs has become a hard barrier preventing their expansion to more application contexts, due to two physical limitations: (1) The massive usage of routing multiplexers causes delay and power overheads as compared to ASICs. To reduce their power consumption, FPGAs have to operate at low supply voltage but sacrifice performance because the transistors drive degrade when working voltage decreases. (2) Using volatile memory technology forces FPGAs to lose configurations when powered off and to be reconfigured at each power on. Resistive Random Access Memories (RRAMs) have strong potentials in overcoming the physical limitations of conventional FPGAs. First of all, RRAMs grant FPGAs non-volatility, enabling FPGAs to be "Normally powered off, Instantly powered on". Second, by combining functionality of memory and pass-gate logic in one unique device, RRAMs can greatly reduce area and delay of routing elements. Third, when RRAMs are embedded into datpaths, the performance of circuits can be independent from their working voltage, beyond the limitations of CMOS circuits. However, researches and development of RRAM-based FPGAs are in their infancy. Most of area and performance predictions were achieved without solid circuit-level simulations and sophisticated Computer Aided Design (CAD) tools, causing the predicted improvements to be less convincing. In this thesis,we present high-performance and low-power RRAM-based FPGAs fromtransistorlevel circuit designs to architecture-level optimizations and CAD tools, using theoretical analysis, industrial electrical simulators and novel CAD tools. We believe that this is the first systematic study in the field, covering: From a circuit design perspective, we propose efficient RRAM-based programming circuits and routing multiplexers through both theoretical analysis and electrical simulations. The proposed 4T(ransitor)1R(RAM) programming structure demonstrates significant improvements in programming current, when compared to most popular 2T1R programming structure. 4T1R-based routingmultiplexer designs are proposed by considering various physical design parasitics, such as intrinsic capacitance of RRAMs and wells doping organization. The proposed 4T1R-based multiplexers outperformbest CMOS implementations significantly in area, delay and power at both nominal and near-Vt regime. From a CAD perspective, we develop a generic FPGA architecture exploration tool, FPGASPICE, modeling a full FPGA fabric with SPICE and Verilog netlists. FPGA-SPICE provides different levels of testbenches and techniques to split large SPICE netlists, in order to obtain better trade-off between simulation time and accuracy. FPGA-SPICE can capture area and power characteristics of SRAM-based and RRAM-based FPGAs more accurately than the currently best analyticalmodels. From an architecture perspective, we propose architecture-level optimizations for RRAMbased FPGAs and quantify their minimumrequirements for RRAM devices. Compared to the best SRAM-based FPGAs, an optimized RRAM-based FPGA architecture brings significant reduction in area, delay and power respectively. In particular, RRAM-based FPGAs operating in the near-Vt regime demonstrate a 5x power improvement without delay overhead as compared to optimized SRAM-based FPGA operating at nominal working voltage.