Top-down analog flows. Myth or reality?
Brian Bailey - May 2, 2012
There are very few designs these days that are not mixed-signal, meaning that analog functionality has been integrated onto the same die as digital logic. This is being done for several reasons, such as reduced cost, but perhaps more importantly, many analog functions now require some digital logic to allow calibration of the analog circuitry.
Traditionally, analog design has been done in a bottom-up manner meaning that the circuit is first designed and verified at the transistor level. It is then integrated with other blocks, also at the transistor level and re-verified. This cannot go on for long before the limits of simulation are reached. If a high-level functional model of the block is required for system-level analysis, a model is built after the fact that matches the implementation. It has been argued for a long time that a top-down methodology, similar to the one used in digital design, would be better. Here the high-level model is created first and the functionality is verified in the context of the system before implementation starts.
Analog designers do not have the same levels of automation available to them compared to digital designers. There are no abstractions that have been created that allow a predictable implementation where concerns such as functionality, timing etc. can be essentially separated and dealt with independently. Back in 1997, Collet International did a study that showed that analog and mixed-signal design takes three to seven times more effort per transistor than digital design. There is no reason to believe that this has changed significantly.
With the introduction of digital logic to help calibrate analog circuitry this may start to force the issue. Several standard languages, such as Verilog-AMS and VHDL-AMS, now exist that enable digital and analog parts of the design to be modeled together. Most simulators now allow a free and efficient mixing of the two, meaning that models at different levels of abstraction can be intermixed and there is no longer a requirement to isolate the digital and analog parts of a design into different blocks. There are also tools that help with some of the back end layout and routing that help to avoid mistakes such as matching. This ensures that devices forming current mirrors or differential pairs remain matched in terms of the parasitics.
My curiosity was peaked when I did a search for users of a top-down analog mixed-signal flow. Most of the papers I found were written by Cadence and there did not appear to be a significant change in the messaging over the past decade. I found evidence of two customers who had worked with Cadence on a top-down flow and there is Cypress who has a programmable device that contains analog building blocks that can be programmed in a high-level manner.
I have two questions. How widely is a top-down analog mixed-signal flow being used in the industry? And secondly, if higher levels of automation existed in the analog flow, would it enable analog designers to create larger blocks, or would they continue to develop the same blocks, and just do it faster or more reliably (assuming this were possible)?