History of CDL

Cyclicity CDL was created by Gavin Stark. Its purpose is to make hardware design more productive, and it comes from a history of software development for ASIC design. In previous projects C models have been created by hand, or perhaps automatically from Verilog (e.g. the RACE simulator acquired by Cadence, a long time ago now). The first course duplicates design work; the latter requires some tool to take a non-cycle based language and convert its output to something cycle based. Neither of these is a good path.

Some vendors have taken the latter path, though, since Cadence dropped RACE. TenisonEDA, for example, now touts VTOC to convert verilog to SystemC, C or C++. This is mainly viewed as a path to create C models from golden verilog, though.

SystemC is then an attempt to create hardware design at a higher level with C models; the major issue with SystemC, though, is the verbosity required to design anything within SystemC, as the C language itself (which SystemC comes under) is not an HDL. Hardware designs are difficult enough to understand on their own, without many layers of extra rubrik obfuscating the design.

SystemVerilog is a good step forward from the Verilog world, and Accellera should be praised for having some insight here. SystemVerilog is a practical EDA solution to the problem, if it is incorporated with a good compiler. However, it suffers from the same problem that Verilog does, in that it is inherently a behavourial test bench language and hardware design language, which means bad design practice easily inflicts pain on hardware designs (more bugs etc) and on compiler writers (far too many constructs are possible, which probably all need to be handled to cater to the whims of designers).

Finally, no good solutions in the compiler realm are available in the public domain. What about low cost hardware design? How do we more readily push the bounds of hardware design practice? Waiting for EDA tool companies is a painful occupation. SystemVerilog is catching up to software design practices of 20 years ago.

So, with time available, and a desire for a good C-model and synthesizable verilog EDA tool, where to go? That was the birth of Cyclicity cycle design language. Take a wealth of programming experience (Verilog, VHDL, functional languages, C, C++, Java, assemblers etc), a wealth of design experience (silicon, board level, FPGA, device driver, OS, tools, application), and focus on a language for ASIC design. Where readability of code is a primary goal; if code is transparent, then bugs are fewer initially. Where good design practice is built in; no unexpected transparent latches, no unreset flops, etc. Where documentation is visible, and can be easily extracted to form supporting release documents. Where hardware design, which is difficult at the best of times, is the focus of the tool.

The obvious starting places are Verilog or VHDL and C. VHDL, with its cumbersome type mechanisms and libraries to do anything, and its ADA-like heritage, is a less good starting point, particularly as almost every low-level designer (hardware or software) needs profficiency in C. Verilog suffers also from a hardware design heritage; it is so focussed on producing the whole environment, that it is easy to miss the wood for the trees in such designs. C suffers also, with no focus on hardware design at all. But a good hybrid of styles, and even syntax, for Verilog and C is workable. This, then, is the basis of Cyclicity CDL.

About the originator

Gavin Stark is a Briton transplanted to California's Bay Area, his life uprooted by the demands of the silicon industry.

He was awarded his PhD by Cambridge University, England, whilst working for his first startup called (at the time) ATML, a company which has since become Virata, Globespan/Virata, and now something to probably become Conexant.

Designing ATML's first system-on-a-chip (i.e. integrated CPU with networking functions) took him to work with Cirrus Logic in Fremont, California, where he eventually took up employment, working as the architect for Cirrus Logic's communications silicon division. This division spun off to become Basis Communications, a privately held communication silicon vendor, of which Gavin was Chief Technical Officer. Here Gavin designed most of the silicon product for Basis, utilizing hand-coded cycle-based C models to do the design, coupled to simple models of an ARM CPU; Gavin also then wrote verilog models using the C models, running both in parallel in simulation to ensure identical functionality.

Upon completion of the design of its first silicon (where the design worked first time), Basis was acquired by Intel Corporation, where it was taken under the wing of the Network Processor Group. Here, Gavin held a post as an architect for the lower end of the network processor line, being chiefly responsible for the architecture of the IXP425 amongst other devices. Gavin left Intel in March 2003.

Main links

SourceForge.net Logo

Site map