with an ASIC design style. The framework is the same; the platforms are different. The number and location of platforms in the design abstractions, the number and type of components that constitute a platform, and the choice of parameters to represent the com- ponents are critical aspects of this method.
As Figure 4 shows, platforms form a stack from the design specification defined in the application domain to implementation. Some platforms demark boundaries critical in the electronics supply chain. These articulation points warrant particular attention. We call an architecture platform the articulation point between system architecture and microarchi- tecture. Microarchitecture is a platform whose components are architectural elements such as microprocessors, memories, and interfaces. This articulation point is where the application engineer maps a design into a physical struc- ture. To find a common semantic domain, the application engineer must abstract these com- ponents via an operating system, device drivers, and a communication mechanism. The hard- ware components support the execution of the specification behavior. Another essential plat- form is the one that corresponds to the layer between design and manufacturing—the implementation platform.
Fellow researchers and I are testing the new methodology in advanced industrial environ- ments. For example, consider an ECU design carried out in collaboration by Parades, Mag- neti-Marelli, and STMicroelectronics.6 This design has a strong control component and tight safety constraints. In addition, the appli- cation contained a large portion of legacy design. The electronic engine control subsys- tem’s functionality comprises the following elements:
failure detection and recovery of input sensors;
computation of engine phase, status, and angle; crankshaft revolution speed; and acceleration;
injection and ignition control law; and
injection and ignition actuation drivers.
The existing implementation had 135,000 lines of C source code without comments.
Our first task was to extract the precise func- tionality from the implementation. To do this, we used a representation based on the Code- sign Finite-State Machine network, the com- putation model used in the Virtual Component Codesign (VCC) development environment,7 resulting in 89 CFSMs and 56 timers. We completely rewrote the behavior of the actuators and some of the sensors in the formal model. For the ignition and injection control law, we encapsulated the legacy C code into 18 CFSMs representing concurrent processes. We redesigned the software to make mapping into different microarchitectures rel- atively easy. In particular, we tested three dif- ferent CPUs and, for each, two different software partitionings, to verify functionality and real-time behavior. In addition, we explored three different architectures for the I/O subsystem: one with a full software imple- mentation, one that used a peripheral (pro- vided by the CPU vendor) for timing functions, and one with a newly designed, highly optimized full-hardware peripheral. Performance estimation based on VCC resulted in an error of only 11 percent, com- pared with a prototype board containing real hardware. We implemented functionality on three platforms, resulting in software reusabil- ity of more than 86 percent. We also used the functionality captured in semiformal terms to design a new dual-processor architecture being
sampled at STMicroelectronics.8
n considering the future of electronic devices and infrastructures as they affect the car manufacturing and design world, the issues related to choice and design of inte- grated platforms are crucial. A rigorous design methodology based on separation of concerns (function and architecture, function and com- munication) and platforms is essential. The challenges ahead are great, but the opportu- nities are enormous. The automobile industry is on the edge of a revolution in the way it I
conceives and designs a car.
I thank the Parades researchers and the Cadence Automotive Tiger Team—in particu- lar, A. Ferrari, M. Chiodo, P. Giusto, L. Lavagno, and J.Y. Brunel—for their contribu- tions to the vision presented here. This research