Model-based design and early verification aid designers
Modeling and simulation at levels ranging from chip to system can catch bugs early and ensure quick time to market.
By Rick Nelson, Editor-in-Chief -- EDN, December 15, 2009
For More Information
Alliance Spacesystems www.alliancespacesystems.com
AppliedMicro www.appliedmicro.com
Argonne National Laboratory www.anl.gov
ARM www.arm.com
Bell Helicopter www.bellhelicopter.com
Carbon Design Systems www.carbondesignsystems.com
CoWare www.coware.com
EcoCar Challenge www.ecocarchallenge.org
Freescale Semiconductor www.freescale.com
General Motors www.gm.com
Hyundai www.hyundai.com
IBS Research and
Consultancy
www.ibsresearch.com
Manroland www.manroland.com
The MathWorks www.mathworks.com
Mentor Graphics www.mentor.com
MIPS www.mips.com
National Aeronautics
and Space Administration
www.nasa.gov
National Instruments www.ni.com
Open SystemC Initiative www.systemc.org
Philips www.philips.com
Rose-Hulman Institute
of Technology
www.rose-hulman.edu
SimuQuest www.simuquest.com
Synopsys www.synopsys.com
US Department of Energy www.energy.gov
Engineers can save themselves a lot of grief if they carefully evaluate their designs in software before committing to silicon or sheet metal. Model-based-design and early-verification tools are helping designers, whether they are developing ICs or airframes, find inevitable mistakes early and get to market on time. Model-based design can be of significant value in helping isolate domain experts, such as medical-device or aerospace engineers, from the need to understand low-level hardware and software details (Reference 1). However, they can also help catch errors at the specification stage that designers would not otherwise catch until the test stage, saving time and money. The available modeling and prototyping tools span the gamut of applications, from mechatronics systems to RTL (register-transfer-level)-IP (intellectual-property)-based SOCs (systems on chips).
Not surprisingly, the automotive industry is aggressively using model-based design and simulation as it grapples with new technologies, such as hybrid-electric and fuel-cell vehicles. A recent article describes automotive-R&D activities as comparable to those of the telecommunications industry when it overcame power and chip-size challenges to support the evolution of cell phones into multimedia devices (Reference 2). The article touts simulation as a way to meet the challenges, citing Mentor Graphics’ SystemVision, for example, as a simulator that can help engineers manage mechanics, electronics, software, and controls all in one system.
Model-based design and rapid prototyping can be particularly valuable in proof-of-concept work or in ensuring that a specification meets customer requirements. Alliance Spacesystems, for example, uses the concept in its development of robotic arms for applications including the Mars Spirit and Opportunity rovers (Reference 3). Sean Dougherty, mechatronics-group supervisor at the company, describes a Hubble-space-telescope application in which replacement of a malfunctioning instrument would require the removal of 100 fasteners. NASA (National Aeronautics and Space Administration), he says, was not sure the feat was possible. Using hardware and software from National Instruments, however, Alliance Spacesystems within three months prototyped a functioning robot with X-, Y-, and Z-axis motion, complete with a vision system to recognize the fasteners.
A more down-to-earth application for Alliance Spacesystems involves automobile-mounted camera booms that the movie industry uses to film car chases. Although aerospace and similar programs tend to have long leadtimes with thoroughly reviewed specifications and requirements documents, filmmakers are more likely to require frequent iterations to obtain what they want. Hardware and software, such as National Instruments’ LabView and CompactRIO platform, says Dougherty, enable him to adapt to customers’ evolving needs, rapidly developing new versions.
Brett Murphy, manager of technical marketing at The MathWorks, lays out the case for early verification. He says that aerospace and automotive members of the company’s customer-advisory boards cite verification and validation as top priorities. Errors most often emerge at a project’s specification stage, and fewer errors manifest themselves at the subsequent design, implementation, and test stages. In contrast, engineers frequently don’t detect the errors until the test stage. Murphy presents a multistage approach to catching and correcting errors earlier. This approach includes capturing requirements using executable specifications; using models as the system-level test benches for algorithms and components; simulating to explore design trade-offs, component interactions, and system-level metrics; and reusing the same test bench from virtual system integration through to the developed system. These techniques are applicable to adopters of model-based design; groups developing control systems; and engineers designing algorithm-intensive signal-processing, imaging, and communications systems.
For adopters of model-based design, Murphy adds, simulations often do not connect to requirements, a problem that designers can address through the definition of requirements-based test in a process that enables the use of simulations to ensure designers find requirements errors early. Companies that have successfully employed the approach include Bell Helicopter. In addition, Murphy says, Hyundai has employed MathWorks and SimuQuest tools to model, simulate, prototype, and deploy an engine-control unit. The MathWorks is also involved in the EcoCar challenge for engineering students (see sidebar “Tomorrow’s engineers learn model-based design”).
Control-system designers, Murphy says, face challenges as system complexity grows. It then becomes important to test control algorithms through modeling and simulation and to leverage models through automatic code generation for a microcontroller, an FPGA, or a programmable-logic controller to support real-time testing. Murphy notes that Manroland has employed MathWorks tools to design and model a printing-press controller, run real-time simulations, and deploy a production system, reducing development time by a year.
For algorithm-intensive signal processing in communications, electronics, semiconductor, imaging, medical, and aerospace applications, verification time and costs are escalating, with engineers spending 50% or more of their time writing verification code. To alleviate the problem, engineers can turn to multidomain system verification, integrating Matlab, C/C++, and HDL (hardware-description-language) IP into Simulink models. Designers can develop a golden reference model in Matlab, develop a test bench in Matlab or Simulink, and perform cosimulation with embedded IDEs (integrated development environments), HDL simulators, or analog simulators, leading to a DSP or an FPGA prototype without requiring low-level programming. Murphy adds that acoustical engineers from Philips, who were not expert programmers, were able to develop and test real-time prototypes for a surround-sound system without writing any low-level DSP code.
Software differentiation
Frank Schirrmeister, director of product marketing for system-level solutions at Synopsys, sees the emergence of the software-differentiated-hardware era as a key challenge, with software-development costs beginning to represent more than 70% of total development costs as process geometries shrink to 22 nm. Citing IBS Research and Consultancy, he says the software costs were less than 20% of total development costs at the 180-nm node. Synopsys’ own research reflects the trend toward higher software costs. As a result, designers are using embedded processors within their designs to verify the hardware—for example, by running test benches.
Synopsys offers a variety of system-level tools, including Saber, System Studio, and Innovator for algorithmic, mechatronic, and system simulation; the DesignWare system-level library of high-level IP; DesignWare cores; System Studio for high-level DSP design and verification; Innovator for virtual prototyping of embedded software; Confirma for rapid prototyping; and core tools for hardware design and verification. One of the company’s newest tools in this market is the Synphony high-level-synthesis tool, which debuted in October. It converts Matlab M scripting-language code to synthesizable datapath-RTL logic.
Synopsys’ DSP-algorithm portfolio enhances verification productivity through HDL-import and SystemC-export capabilities and maximizes simulation performance. System Studio supports algorithm optimization and verification using an executable test bench. The DesignWare system-level library combines with Innovator to support early software development and enhance design quality through a SystemC-executable specification.
Cycle-accurate modeling
Carbon Design Systems focuses on cycle-accurate IP modeling for virtual prototyping of SOCs, providing presilicon validation of hardware and software designs. The company offers a compiler that reads in RTL in VHDL (very-high-speed hardware-description language) or Verilog and produces a model that you can link into almost any virtual platform, including Carbon SOC Designer, CoWare Platform Architect, and OSCI (Open SystemC Initiative) SystemC (Reference 4). Cycle-accurate models in a language such as C represent the RTL from which designers compile them and consequently are more suitable for SOC validation than are models you derive from behavioral descriptions, which may not match the RTL performance.
|
According to Bill Neifert, Carbon’s vice president of business development, the company has recently been working with ARM and MIPS Technologies on cycle-accurate models for increasingly complex processors. “Continuing handwriting cycle-accurate models was becoming too onerous a task [for ARM] as the company developed more and more advanced processors,” he says. “It turned out it was just as much work to write the cycle-accurate models as it was to write the RTL for the real design.” As a result, ARM decided 18 months ago to stop handwriting models. “That’s when we stepped in and said that model generation happens to be a problem that we’ve solved,” says Neifert. Carbon then acquired ARM’s SOC Designer tool, which handles model generation for ARM IP, generating 100%-accurate models for ARM RTL IP and integrating features such as debuggers and program loaders. MIPS faced similar challenges in handwriting cycle-accurate models. “Seeing what we are able to do with the ARM processors, MIPS decided to follow a similar path,” Neifert adds. MIPS, however, chose to employ Carbon’s technology to generate its own cycle-accurate models of IP, such as its M14K and M14Kc cores, for its customers.
IP providers such as ARM and MIPS aren’t the only customers for Carbon’s technology. Last month, Carbon announced that AppliedMicro, which addresses energy-conscious computing and communications applications, selected Carbon Model Studio to accelerate the deployment of SystemC-based virtual platforms for presilicon and postsilicon software development, performance analysis, and validation of SOC designs.
| References |
|
| Tomorrow’s engineers learn model-based design | ||
|
The engineers of tomorrow are getting a taste of model-based design and simulation through programs such as EcoCar: The Next Challenge, which has the support of the US Department of Energy, General Motors, The MathWorks, Freescale Semiconductor, National Instruments, and other sponsors that are lending time, money, and products to the university-level competition. The EcoCar program began in the fall of 2008 when students from 17 universities in the United States and Canada began a three-year effort to design and re-engineer a 2009 Saturn Vue to be more efficient and reduce emissions (reference A and referenceB). Speaking at a kickoff meeting for the second year of the program, which took place at The MathWorks headquarters in Natick, MA, Mike Wahlstrom, a control and simulation engineer at the Center for Transportation Research at the Energy Systems Division of Argonne National Laboratory, said that EcoCar is the latest effort in the 20-year history of advanced-vehicle-technology competitions. Wahlstrom began his career with the competitions as a student and now assists in organizing the program, which Argonne manages for the Department of Energy. For the first year of the program, teams worked only on model-based design and simulation. Vehicles began arriving this fall, and teams will spend the second year building a “mule vehicle,” a term the teams borrowed from GM’s global-development program. A mule vehicle will be 60 to 65% ready to go. During the final year, teams will refine their vehicles to bring them nearer a production state. Chad Conway, a sophomore at Rose-Hulman Institute of Technology, joined the team last winter and has been using MathWorks and National Instruments tools to investigate vehicle design. Zachery Chambers, an associate professor of mechanical engineering at the school, says that the team is working on a parallel pretransmission/post-transmission electric hybrid. The vehicle will have a downsized 1.3l diesel engine. Between it and a stock four-speed automatic transmission, the team will insert an electric machine to provide additional capability to the engine to offset its downsizing. In addition, the team will place another electric machine on the rear axle to provide for regenerative braking and additional acceleration. As for the importance of model-based design and simulation, Chambers says, the biggest mistake teams can make is to take a wrench to their vehicles too early. “When they have an operational vehicle, it really behooves teams to use their CAN [controller-area-network]-analysis tools to tap into the vehicle network to make sure they can communicate with all the modules inside the vehicle and go out and collect some baseline data.” Chambers advises the teams to figure out their testing procedure now while they have something that works before they try to “figure out a testing procedure with something that may or may not work all that well.”
|
