Building an embedded product is a complex undertaking, involving highly intertwined hardware and software components and thousands of individual design decisions. The pressure to deliver a product to market quickly makes complexity still more difficult to manage, because design trade-offs and alternatives can't always be properly tested and evaluated.
Simulation using software prototypes (models) of the hardware design is one path to establishing a design direction and making intelligent trade-off decisions. Traditionally, however, simulation has suffered from poor performance (slow execution times) and inadequate information. While system-level simulation has been useful in modeling and testing a particular component or interaction, under most circumstances the technique has not been able to deliver the level of detail needed to manage complexity and thereby accelerate product delivery.
Nonetheless, simulation techniques have come a long way. Today's virtual prototypes can handle entire systems, modeling components, signals and hardware-software interactions with a high level of accuracy. And if you're willing to accept trade-offs between the level of detail and level of performance, you can have a bit of both.
The result is a new product design flow that emphasizes hardware and software co-development based on a design prototype.
How is that done? First, commercial prototype intellectual-property libraries, such as those available from Synopsys Inc. and Vast Systems Technology Corp., have become effective in building detailed software models that reflect popular processors, components and even entire reference designs. In many cases, these software models are available early in the product life cycle, sometimes even before silicon is available.
Second, the performance of simulated systems has improved enormously. That has been achieved in part by the faster and more-efficient processors employed by the latest desktop engineering systems. Additional performance improvements come from more-intelligent simulation systems.
The ability to simulate critical components with a high degree of accuracy, combined with the early availability of such simulation models, offers the potential to accelerate product design and even transform the design process for many embedded systems.
Market-driven trend
As the ability to prototype and simulate embedded systems improves, more designers are adopting the technology to speed product development. Vendors such as Vast Systems, Synopsys and Paravirtual Corp. are finding fertile ground in design projects that combine complex hardware and software, yet are highly market-driven.
According to Jeff Roane, director of product marketing for Vast Systems, certain markets have adopted virtual prototypes as an integral part of the design process. "Wireless, consumer and automotive have all been leaders in using virtual prototypes. They all have common characteristics," Roane said.
The most important characteristic is time-to-market. The markets Roane cited are all segments in which millions of dollars can be made or lost in a matter of weeks, depending on the availability of cutting-edge products.
Larger semiconductor design vendors also see value in virtual prototypes. Earlier in the year, Synopsys acquired Virtio, another leader in software for design prototypes. In this case, "it's all about intellectual property," said Filip Thoen, senior staff engineer for R&D at the Synopsys System Level Solutions Group. "We can leverage the Synopsys library of IP to offer a comprehensive solution for many different design projects."
Prototypes combine highly detailed and accurate models of many hardware components. As such, they represent IP almost as much as the hardware that they model.
The combination of Synopsys' System Studio with Virtio's virtual-prototyping technology gives software developers the ability to begin software development much earlier than usual. The acquisition also enables Synopsys to provide an integrated implementation, verification and IP solution to speed hardware and software development.
The key to virtual-prototype adoption and use is cycle-accurate simulation, which provides for each hardware cycle to produce exactly the same outputs as the actual hardware would. Because cycle-accurate simulations are specific to the modeled hardware, they must be adapted to each component for which design engineers require a prototype.
Vendors of virtual-prototype software have to work with hardware providers to create models of parts before those products become generally available. For example, Virtio has in the past worked closely with Texas Instruments Inc. on the development of prototypes for wireless platforms, such as TI's Omap.
The models are typically made available at the same time as, or even before, the actual silicon platform, making it possible to begin system validation and software development early.
Of course, not every operation is replicated in the prototype exactly as it is run on the hardware. For a deterministic activity, the prototypes will typically return the expected value, without necessarily going through all of the actual computations. That improves simulation performance without affecting the results.
Additionally, designers can usually mix highly detailed, cycle-accurate simulation with less-detailed, cycle-approximate simulation, thereby focusing on the most-critical aspects of the hardware. In practice, an executing simulation is driven by the cycle-accurate timings, with appropriate handoffs to the less-detailed portions.
In all other aspects, the prototype can behave exactly as the hardware does, even to the point of running not only application software but also a real-time operating system (RTOS). Virtual prototypes can support several popular commercial RTOSes and any software that runs on top of those operating systems. While it is unlikely the virtual prototypes will perform as fast as the actual hardware, their timing will be accurate.
But as designs gain in complexity, especially with systems-on-chip that employ multicore architectures, virtual prototypes are stressed to keep up. Prototypes not only must combine and keep track of the timings of multiple core models, but they also must maintain simulation performance as cores are added. The goal is linear performance degradation as the virtual prototypes add cores.
Vast Systems claims to have achieved that goal with its recently released Comet6 development environment. The company has also taken the Eclipse plunge, joining other embedded companies in hosting development tools in this extensible platform. The use of Eclipse makes it possible to create a custom environment for building and running code on prototypes.
New processes at hand?
The move toward virtual prototypes is occurring not only because of the improvements in technology and performance, but also because traditional design processes are hitting a wall. The old methodology--producing a design, building hardware to that design specification, writing software and delivering the resulting product--is costly and time-consuming. Further, if there is a problem with the design, the problem may not be found until software is tested. That can delay final implementation.
