Synchronizing Designs and Behavioral Models in Mixed-Signal Flows

The creation of behavioral models is only one part of the process of using those models in a mixed-signal design verification flow. If the model and design don't match, the effort is worthless. Even worse, it can damage the entire design verification process.

"Why should I care about keeping my behavioral models and designs in synch ?"

The benefit of using a bottom-up behavioral model to improve design verification simulation performance is totally wasted if the model being used does not match the behavior of the design which it replaces. In fact, a bad out of synch model can expose verification issues that don't really exist, or hide those that do.

"How can I ensure my behavioral model and design have equivalent behavior ?"

Until recently, this required a lot of manual effort and good discipline (pardon the pun):

• Recognize when the design or model interfaces or internals have changed.
• Modify the model (bottom-up) or the design (top-down) as required.
• Modify the test bench to match the new behavior, pins, etc.
• Simulate the design and model over various swept parameter ranges.
• Compare the measured behaviors from simulating the design and model using the modified test benches.
• Manually eyeball the overlaid design and model waveforms to ensure that no unacceptable differences, glitches, timing skew, etc. problems have appeared.

This can take many hours of manual effort when and if the design or model change is recognized. The cost later in the design flow if these differences are not recognized can be several orders of magnitude higher.

"I don't have time to do all that, what can you do to help ?"

Based on the feedback and requirements from many customers who are actively using behavior model models within the their design verification flows, we developed amsDmv (Analog Mixed Signal Design and Model Verification).

amsDmv is fully integrated with Virtuoso, ADE L/XL, ViVa and SimVision to provide a complete self contained model verification and debugging solution. Features include:

• Interactive setup of model vs. design verification requirements
• Simulation of design and model using ADE L, ADE XL, OCEAN, irun, etc.
• Verification of design vs. model ADE L and ADE XL waveform signals
• Verification of design vs. model interfaces (pin names, order, direction, etc.)
• Interactive debugging of verification failures including waveform zoom to failing areas
• Save of entire setup to a UNIX batch command for inclusion in nightly regression runs

The amsDmv flow is shown below. 

 

 

 

The user provided design, model and test benches are simulated using the relevant tools and flows within amsDmv to create the required measured result behavior and waveform signals. As selected and set up by the user, amsDmv validates: 

• The design and model measured results to user provided tolerances.
• The design and model waveforms to user provided tolerances.
• Selected different elements of the interface and pins of the model and design.

The waveform validation algorithms are especially powerful and useful because for many design types, the user does not need to create specific measurements in the test benches. The algorithms highlight unacceptable absolute and relative amplitude and time differences. The amsDmv design vs. model validation, once interactively set up, can be exported to a UNIX batch script that can be run within a nightly regression run.

If the differences between the design and model exceed the user provided tolerances, the batch script will fail and report problem areas that can later be debugged interactively within amsDmv.

The need for design vs. model validation within a complete mixed signal design flow

Whether the model is created manually or via Schematic Model Generation (SMG, to be covered in a later Blog), there are many checkpoints within the design flow where it is important that the model and design are checked to be equivalent. As the design progresses, the design or model may need tweaking, or the model may need to be recalibrated to measurements made by simulating the post-layout extracted view with parasitics.

 

 

 

The above picture shows a top-down and bottom-up mixed signal design flow. A top-down model is created to encapsulate the required specifications in a simulatable representation. The transistor level implementation can then use the top-down model as a guide for the designs requirements and behavior.

Once the design has been through layout and the parasitics extracted, the post layout design can be simulated and the model recalibrated to create a silicon calibrated behavioral model that represents the behavior of the finished design. At each of these points in the design flow, the different abstractions and phases of design and model creation can be cross validated to ensure consistency.

Summary

The onerous time intensive and error prone task of ensuring that behavioral models and designs stay in synch can be significantly improved and automated by using amsDmv. This  enables users to not only improve design verification simulation performance, but also to trust that the results they are seeing do accurately match the real design.

Paul Foster

(CIC / CSV Architect)