1 / 13

Platform-Based Design A paper by Alberto Sangiovanni-Vincentelli EE 249, 11/5/2002 Presenter:

Platform-Based Design A paper by Alberto Sangiovanni-Vincentelli EE 249, 11/5/2002 Presenter: Mel Tsai. Outline. The Goals of Design The Problem & Trends ASV’s Solution: Platform-based Design Why does it work? Other Definitions of “Platform” The System Platform Stack

Download Presentation

Platform-Based Design A paper by Alberto Sangiovanni-Vincentelli EE 249, 11/5/2002 Presenter:

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Platform-Based Design A paper by AlbertoSangiovanni-Vincentelli EE 249, 11/5/2002 Presenter: Mel Tsai

  2. Outline • The Goals of Design • The Problem & Trends • ASV’s Solution: Platform-based Design • Why does it work? • Other Definitions of “Platform” • The System Platform Stack • Architecture Platforms • Software (API) Platforms • Other Platforms • Conclusion

  3. The Goals of Design • Minimize: • Development costs • Production time • While Maximizing: • Performance (i.e. speed, low power consumption) • Functionality (i.e. more features) • Production volume (i.e. more profit)

  4. The Problem • Development & production costs are expensive • Chip geometries are shrinking – the billion-transistor-on-a-chip age. • Complex chips have long design cycles & difficult verification issues… There is a designer “productivity gap” that is exacerbated by short time-to-market. • High NRE costs: Mask sets & probe cards are becoming prohibitively expensive. ITRS predicts >$1M for 0.15 micron and smaller.

  5. Some Trends • Full-custom design is being replaced by “library-based” design • Trend towards design at higher levels of abstraction • Trend towards hardware re-use, in both space and time • Short TTM requires architectures that can respond to late specification changes • Trend towards architectures with HW or SW programmability • Fewer design starts • Trend towards higher-volume production of fewer chips

  6. ASV’s Solution: Platform-based Design • Basic idea: • Goal is to hide the details of a design through a layered system of abstractions. • Identify the most natural and important layers of abstraction, and call them platforms. • Application Programming Interface (API) platform • Architecture platform • Silicon implementation platform • Design by successively refininga platform instance until it becomes an instance of a subsequent (lower) platform.

  7. Application Application Silicon Implementation Silicon Implementation Why does it work? • Design time & cost are minimized: • Design-space at each layer is restricted because underlying layers are abstracted-away. Hence, the design of each platform instance is more manageable. • Each layer can be (mostly) designed independently of the others, eliminating the need for complete design-loop iterations when things go wrong. • Hardware and/or software components can be re-used if they can be folded into a platform and abstracted properly.

  8. Other “Platform” Definitions • Many overlapping & outdated definitions in the hardwarecontext: • A flexible integrated circuit that is customized via programming • A family of chips, based on the same microprocessor, that support various applications with different performance requirements • A family of architectures that all support the same characteristic abstraction, i.e. the “PC platform” • A family of architectures that satisfy a set of architectural constraints imposed to allow the re-use of hardware and software components.

  9. The System-Platform Stack Application Space • System-Platform Stack: the “sandwich” consisting of the API platform and the Architecture platform. • ASV’s Platform-based design is a “meet-in-the-middle” approach • It is top-down because constraints and functionality propagate to each lower platform instance • It is bottom-up because both the characteristics and performance of the lower levels must be exported to the upper platform Application Instance Platform Specification API Platform Architecture Platform Platform Design-Space Exploration Architecture Instance Architecture Space

  10. Architecture Platform • In this approach, the architecture platform should support architectures that are “mainly geared towards optimizing design-time.” • This implies an approach based on a family of similar micro-architectures, i.e. programmable architectures, that can be extended or reduced for a range of different applications. • There is a trade-off between flexibility of the architecture platform and design-time. • Fewer constraints on the platform provide better support for applications & optimization, but constrained architectures have fewer parameters and are easier to assemble & verify.

  11. Software (API) Platform • Software development is now 80% of the design of embedded systems • The re-use of application software is a must • The goal of the API platform is to present an abstracted interface to any proposed instance of the architecture platform. • The API platform is a software interface that wraps: • The processor & memory via an RTOS • Device drivers and I/O • The communication network

  12. Other Platform Stacks Architecture Platform Silicon Implementation Platform Stack Silicon Implementation Platform Manufacturing Interface Platform Stack Manufacturing Interface Platform

  13. Conclusion • Platform-based design minimizes design time and cost by breaking the design process into manageable parts • The exploration space for each step is small • Due to the restricted design space, this approach may not produce the highest-performing designs • The validity of this approach depends on the ability to accurately abstract and estimate final performance at each level • Poor estimates and abstractions early-on can be magnified at the later stages

More Related