Evolution and Synthesis

If you have not yet seen it, Jim Hogan and Paul McLellan wrote a great piece over at EE Times entitled "The evolution of design methodology" (part 1).

Their conclusion is that the chip design industry is in the midst of another major shift to one where chip design becomes software-centric. In other words, system houses define the end-product, and much of the differentiation now comes from software. This software lives on across multiple generations of SoC's. So they are building SoC's so that they run their software most optimally for the desired system spec (cost, performance, power consumption, etc). The prototypical example is the A4 chip that Apple designed to power the iPad. The money quote is "the application drives the hardware."

The article also mentions that in order to design algorithms that run the software load optimally, "a lot of the algorithms that will be implemented in silicon (as opposed to running on a microprocessor or digital signal processor) are written in C or C++ and not RTL."

Given the size and complexity of designing and verifying these SoC's, there is necessarily a lot of re-use of blocks and subsystems. This also lends itself to designing in C or C++ and not RTL, since RTL captures the micro-architecture details associated with the target hardware. As a designer, you would structure your RTL differently depending on whether performance, power, or area was the primary concern, and even the target process technology would influence the RTL.

Jim and Paul have always had a way of distilling complex concepts into message that are clear to me, and this is no exception. In order to enable software-centric hardware design, we are going to have to evolve hardware design from RTL to SystemC. To make that work in the real world, we will need High-Level Synthesis (HLS) tools that enable entire blocks and subsystems to be specified in  SystemC and synthesized to RTL that can be implemented through production RTL-GDSII systems and achieve power, performance, and area that is comparable to hand-crafted RTL.

Jack Erickson

Over the past few months, I too have been evolving (my wife says "finally") -- from RTL synthesis to HLS. I have been working with the C-to-Silicon Compiler team, and I'm seeing customers achieve the promise of production-worthy HLS (see references below). Now we need to help support the industry make this large-scale shift. So stay tuned to this space!

Cadence and Renesas Enhance Productivity with New System-Level Design Approach

Casio Sets New Productivity Benchmarks with TLM-Drive Design and Verification

A user's first look at Cadence C-to-Silicon

Adoption of Cadence C-to-Silicon Compiler Accelerates in Japan