Statistics

We have 5 guests and 1 member online

SVTechie Recommends

powered_by.png, 1 kB
System-level design languages: to C or not to C? PDF Print E-mail
del.icio.us It | Furl It | Digg It
Written by SVTechie   

reply.gifOriginal Article can be read at System-level design languages: to C or not to C?
By Graham Prophet, Editor, EDN Europe -- EDN, 10/14/1999

First the bad news: If you have only recently begun to fully exploit HDL-based design with VHDL or Verilog, it will soon be time to move on and move up to a higher level of abstraction. But the good news is that the languages you'll be using may not be entirely new, and they might even be very familiar. 

For as long as EDA has existed, its innovators have had their sights on a single high-level-design environment. In such an environment, you could model the overall behavior of a system before you detail its design and long before you build a prototype. You could plan and explore which areas of your design you could best realize in hardware and which you could best implement in software. And you could use the description you created during the detail-design stages of both the system hardware and the software, constantly refining the system and moving toward a finished product.

Analysis
By SVTechie on September 23, 2005

This articles provides great insight into various tradeoffs between High Level Synthesis Methodologies and is almost 6 year old. But debate is still ON and there is great uncertainty about future of high level synthesis. None of previous high-level synthesis products have been successful. Not surprising because designers have to look at myriad of issues before selecting any methodologies and none of the products was able to address various issues successfully. Designer has to look at Quality of Result, Design Time, Productivity, Cost and in the end, it comes down to compromising on one factor to achieve another.

There have been multiple major reasons for the anemic revenue of previous Behavioral Synthesis companies. One is changes to the C/C++ language to support hardware design and resulted in long learning curve. Even though going to higher level of abstraction help achieve time to market, promised productivity was not delivered. Main reason behind this major productivity loss is incorporation of timing constructs in High level language. Low level timing had to be captured in high level implementation to represent underlying hardware and it was not appreciated by designers. Also synthesized hardware was not able to achieve required performance because of lack of maturity in new mapping technology.

Thus, it took a long time to learn AND SUPPORT the special pseudo-C language and then after all this preparation, the results were poor. So, even though time might be saved in the actual synthesis process (Behavior synthesis's major claim), the preparation time, the additional required exploration step, and the poor results made the traditional ASIC approach much more attractive, and productive.

Following is the outcomes of the C language design companies mentioned in the article. They were also greatly effected by too little money, over-hyped and poor management.

  • Frontier Design - Acquired by Mentor. Not sure if Mentor CatapultC technology is derivative of Frontier Design Technology.

  • IMEC - Excellent European EDA research University. One of the first to successful use C for hardware design and still have high level of research in High Level Synthesis.

  • C-Level - Bane of any High Level Synthesis Company. Over hyped the space and was killed by its' own design. (Fire Sale to Synopsys)

  • CynApps - Merged with Chronology to become Forte Design, which is still in business.


Has anything changed? Yes, a lot and a lot more is desired. It was clear from a discussion with a high level synthesis tool company that making various design tradeoffs is still not possible. There has been scenario in which performance was 2X and area was 4X.

We will look at how market landscape has changed in past years later in another article. Following questions will be investigated later. Has there been a fundamental design/business change that now makes C to hardware a viable business? What was wrong with previous generation of High Level Synthesis Products? (Link will be added here once that article is ready).

This article can be discussed at forum.


Disclaimer, the evil necessity: Posted views are of author only and this website is no way responsible for any damages caused by usage of this information.
 
Last Updated ( Friday, 18 April 2008 )
 
< Prev   Next >
[ Back ]