EDA Tools for Analog: Where Do I Go From Here?

　　Just as analog IC design is evolving, so, too, are electronic design automation (EDA) tools as they evolve to keep up with the demanding verification needs of next-generation chips. However, while analog, mixed-signal, and RF design tools have continued to grow rapidly and have hit double-digit annual growth rates in recent years, they have not exploded in scope to parallel the range of tools for digital design.

　　“The key enabler of digital design automation has been the ability to use abstracted representations of standardized electronic components to synthesize and simulate designs,” said Laurie Balch, research director at Pedestal Research. “This is a well-established practice for digital design, but far more difficult for analog.” That’s because, by definition, analog operations cannot be represented as just zeros and ones, which permits greater design flexibility but also means greater analysis intricacy.

　　Therefore, the EDA industry has not yet successfully developed adequate ways to achieve higher levels of abstraction for analog design techniques. “On top of these technical challenges, there remains both a real and perceived mystique surrounding the artistic element of analog design expertise,” Balch said. She added that analog engineers maintain specialized skills and knowledge to build custom circuitry with minimal standardized components.

　　As a result, automating all the specialized experience, analysis requirements, and tricks and rules of thumb for making design tradeoffs is neither technically straightforward nor readily welcomed by the design community. Moreover, adopting new analog automation tools, even if they can be optimized for excellent performance, will require engineers to shift their mindset and trust tools that promise to offload more of the manual design tweaking and optimization they’re accustomed to conducting themselves.

　　However, Sathish Balasubramanian, head of product, marketing and business development for the AMS division of Siemens EDA, sees some recognition of the advantages of a top-down digital methodology. “There is a paradigm shift underway to adopt digital verification techniques for the functional verification of analog and mixed-signal designs.”

　　Balch also sees some degree of catching up with digital tools in the future. “We fully expect that eventually analog design tools will further mimic the landscape for digital design tools,” she added. “With the ever-increasing analog content embedded in the modern electronic devices, it’s simply not feasible for analog engineers to continue doing so much manual design work.”

　　A modest progress

　　Despite the challenges outlined above, there are signs of progress. Take the case of analog simulators, which must constantly enhance their model parsers to support the latest and greatest process nodes. “This is critical because analog simulators are used to characterize standard cell libraries, which will become foundational digital building blocks for new chips,” Balasubramanian said.

　　He added that the matrix solver is the dominant component of the analog simulator, especially for large circuits, and it’s invoked repeatedly during the simulation. “New algorithms are being incorporated to improve matrix solving, as well as for parallelization, which can reduce the runtimes in circuit simulators.”

　　Analog chip developers—users of these tools—are also expressing a sense of optimism. “Offering lab-quality results for virtual analog designs through EDA tools can mean vast computing power and simulation times,” said Henri Sino, product director of customer tools experience at Analog Devices. “To address this challenge, Analog Devices is prioritizing digitization of go-to-market engineering deliverables, such as datasheets to leverage and scale our EDA roadmap.” He added that Analog Devices is leveraging web-based tools, interactive content, and complete system designs as starting points.

　　Will machine learning matter?

　　When it comes to key challenges and potential solutions, Balch pointed to another vital premise. In the digital design world, increasing design size and complexity using advanced process nodes and materials necessitates more design automation. However, there aren’t enough analog design experts available, and design timelines are too tight for the traditional approaches to continue being sustainable.

　　“It’s entirely possible that machine-learning algorithms may be a key to jumpstarting new automation options for analog design methodologies,” Balch said.

　　Balasubramanian shared a similar view regarding machine learning’s potential in analog EDA tools.

　　“Analog design is no longer restricted to block-level designs like op amps, data converters, and filters,” he said. “So it’s now finding wider applications in artificial intelligence, as analog is a closer representation of how the brain operates.” Balasubramanian pointed out that analog simulation produces a huge amount of measurement data. Here, advancements in machine learning can turn mountains of this raw data into valuable design insight that can improve a designer’s productivity.

　　Not only design data but data associated with the variability of physical attributes can be utilized by machine learning to build variability models. When used for design variability analysis, it can result in 1,000× fewer simulation runs than what is needed by brute-force methods.

　　Analog at the heart of the SoC

　　Although digital circuits are largely responsible for everyday computing and are at the heart of modern chips, analog circuits are still integral to the successful operation of systems-on-chip (SoCs). Take the clock, for instance, the heartbeat of the SoC, sourced from a phase-locked loop, which is primarily an analog and mixed-signal design.

　　Balch summed up the progress by noting that recent developments in analog EDA have largely revolved around better modeling and analysis of the parasitic effects of analog circuitry. Siemens EDA’s mPower tools are a good example in this regard. “Analysis tools and design optimization are certainly critical elements to ensure analog design success, but they’re only part of the long-term vision for analog design automation.”

　　Balch recounted that it was the late 1990s and early 2000s when we last saw an earnest attempt to introduce analog synthesis and abstraction techniques. But these efforts were ultimately unsuccessful. It’s quite possible that the time is now to reinvigorate such approaches using the latest machine-learning techniques. “But it’s a near certainty that analog design methodologies won’t catch up with digital methodologies anytime soon,” she concluded.