Mixed Signal Design IP Embraces Metric-Driven Verification Using RNM

Even though it's been over 2 months since this year's Design Automation Conference in San Francisco, I am still surprised by the response that metric-driven, mixed-signal verification gets from our design community. Cadence had quite a few customer presentations at the EDA360 Theater at DAC this year. However, there was one presentation titled "Metric Driven Verification Approach for Analog/Mixed Signal IPs" authored by Pierluigi Daglio and Marco Carlini from STMicroelectronics that has garnered a lot of interest from the verification community.

Metric-driven verification is the norm for digital designs. But, we can extend this concept to analog/mixed signal designs. Analog/mixed signal verification in the context of full chip verification can achieve a respectable coverage level without compromising on performance levels of digital verification. This can be accomplished by using more robust and abstract analog behavior models such as Real Number (RNM) models using Verilog-AMS wreal as an example. RNM models are also provided in VHDL and System Verilog extensions as well. 

Typically, analog verification is based on detailed transistor level simulations run using SPICE-level simulators in a bottom-up flow. Digital verification is based on a top-down approach and has a uniform verification plan using the Universal Verification Methodology (UVM). This metric driven approach uses coverage-directed random stimulus generation and supports multiple verification languages. However, with the trend towards complex mixed-signal SoCs, analog and digital verification cannot afford to stay isolated from one another.

To address the growing verification challenges in today's mixed signal designs, engineers have been using mixed-signal co-simulation. With this approach, the analog behavior models are modeled using Verilog-AMS or VHDL-AMS languages. Mixed-signal simulators are available to then simulate the analog portion using the analog solver and digital portion using a digital event driven simulation engine. But, this approach has 2 key disadvantages.

• 1. The performance bottleneck of this co-simulation is still governed by slow transistor level analog simulation models. In today's complex digital-centric mixed-signal SoCs, achieving satisfactory coverage level in a reasonable amount of time is an impossible task.
• 2. The expertise required to write efficient analog behavioral model is hard to come by. A badly written behavioral model can cause huge performance degradation.

Newer approaches to using RNM to model continuous analog behavior in discrete digital models are gaining a lot of traction. The wreal language extension in Verilog-AMS offers the best trade-off between performance and accuracy, and thus helps analog designers achieve acceptable coverage levels. This performance gain is achieved by using fast digital simulators like Incisive Enterprise Simulator to replace extremely slow mixed-signal solutions.

In addition to performance gain, RNM models introduce a metric-driven verification approach as well as assertions to analog/ mixed-signal designs. With assertion-based verification, top-level verification coverage levels can be managed to the specifications as desired.

Cadence has further extended wreal beyond the Language Reference Manual limitations for more effective usage, and come up with flow that is unique and efficient. Apart from supporting wreal extensions in its verification offerings, Cadence is now pioneering key technologies that will make the use of RNM adoption easier for the verification teams.

Here are two new methodologies that are needed to accelerate the metric-driven verification approach for mixed signal designs:

• To overcome the inefficiencies in creating models, the process of generating RNM models from schematic design input needs to be automated. This addresses the very important problem of writing RNM models for analog portions.
• A streamlined flow is needed to validate the wreal models against the representative analog design. This automated flow will compare the simulation results from RNM and analog design (transistor-level) and provide pass/fail statistics. Once verified, these RNM models are then qualified to be used in a full chip verification flow.

A typical mixed-signal verification flow using RNM consists of the following:

• Automated RNM model generation from schematics
• Automated RNM model validation using a streamlined approach.
• Metric-driven verification approach using Incisive Enterprise Planner and Verification Cockpit
• Coverage closure and overall verification plan management using Specman/Specman-AMS

To learn more about this topic, please see the following whitepapers:

Solutions for Mixed-Signal SoC Verification

Mixed-Signal Design Challenges and Requirements

You can listen to an audio recording of the STMicroelectronics EDA360 Theater presentation (June 5, 4:00 pm) and view the slides here.

Sathishkumar Balasubramanian