A look at what role the Mentor acquisition plays in Siemens, a year after the deal
SAN FRANCISCO — Siemens closed the acquisition of Mentor Graphics in March 2017. How’s that working out for the electronics industry?
This marriage — an EDA business bought by a powerhouse of infrastructure solutions — initially inspired skepticism among investors, the EDA community and media. However, one year later, speaking of Siemens’ “Digital Factory” strategy in an investors’ conference in New York last month, Siemens AG CEO and President Joe Kaeser laid out facts and made a compelling case for the Mentor acquisition:
For those of you who have been following Siemens for a long time, I know you are questioning time and time again, ‘is it really worth doing this digital factory thing? Is it really coming? Or is it another over-valued, over-rated, waste of money by acquiring companies like Unigraphics, last but not least even Mentor Graphics?’
Ladies and gentlemen, I am here to tell you it’s been the right thing. Actually, it turned out much better than we originally thought. It’s a real thing.
Siemens’ digital factory initiative is paying off nicely, according to Kaeser, particularly after the Mentor deal. Siemens’ 4 percent share in the digital factory market in the first quarter of fiscal 2017 jumped to 20 percent in the second quarter of fiscal 2018.
In an interview with EE Times at Design Automation Conference here this week, Wally Rhines, Mentor’s chairman and CEO, said Siemens PLM/Mentor had “a new record year this past year.” Tony Hemmelgarn, who became the president and CEO of Siemens PLM Software last October, added: “We’re on track to meet really ambitions financial targets for the next year. We stated that we expect to achieve about 3.4 billion euro in revenue and we are well on track to outperform the market.”
Physical world meets virtual world
As Kaeser sees it, Siemens’ acquisition of Mentor was designed to prove that the “physical world and mechanical worlds can be simulated.”
It took Siemens a long time to understand that it’s not just the mechanical world that needs to be simulated, Kaeser acknowledged. The fusion of the industries enabled the miniaturization of electrical systems, and “if you go about miniaturizing systems, you go into semiconductors,” he said. Simulations of miniaturized systems thus demand simulations of semiconductors, and that’s where Mentor comes in.
Hemmelgarn, echoing Kaeser, reiterated that the Siemens-Mentor deal was about “making the best possible virtual representation of a physical world.” The process, also known as “digital twin,” helps customers understand and predict the physical counterpart’s performance characteristics before a system gets built.
As product complexity grows with the rise of software, “it has become much more difficult for companies to validate, prove and build products,” Hemmelgarn observed. One way to quickly respond to the industry’s disruptions is to “leverage the digital environment.” Companies must establish a process to “test and prove out” designs [of the product and production systems] in the virtual world, before investing in physical prototypes and assets, he explained.
Good to their word
Meanwhile, Siemens has been “good to their word,” said Rhines. The parent company invested in R&D staff and made a series of acquisitions over the last 12 months, bringing in companies with key technologies — ranging from 5G and analog designs to machine learning and ISO26262 verification. Retention of employees at Mentor has been high, noted Rhines, while the head count of the company’s R&D staff increased about 22 percent. Among the various IC design divisions, growth has ranged from 15 to 35 percent since Mentor was acquired by Siemens.
The supporting cast acquired by Siemens includes Austemper Design Systems, a provider of innovative IC functional safety technology acquired last week, and Sarokal Test Systems, whose technology allows customers to validate 4G and 5G designs during pre-and post-silicon testing. Also, Solido Design Automation provides machine learning-based variation-aware design and characterization software, and Infolytica adds software and domain expertise for low-frequency simulation to support electric motors, generators and electromagnetic device designs. Finally, Tass International, a vendor of simulation software, also provides the automotive industry with engineering and test services.
Rooted in systems
In a new era with Apple as one of the largest chip designers in the world, “System guys are the next generation IC designers,” said Rhines.
A big plus in this environment is Mentor’s roots in systems’ business, said Rhines. Since the 1980s, when Mentor had Daisy software running on Apollo Computer workstations, Mentor’s tools were already in place at various systems companies, and have stayed there for a long time.
Combined with Siemens’ heritage in automotive and aerospace businesses (Siemens covers everything from systems designs to manufacturing and product lifecycle management), Mentor has gotten a head start in a systems-centric, brave new world where every EDA vendor is racing to win.
Asked how an EDA company must do business differently when working with chip designers at systems companies, Rhines said, “In general, a systems group is always very concerned about systems requirements.” Tools capable of “requirements tracing” are critical, he said. A systems group wants to make sure that any designs relative to wiring, signal integrity or cost optimization can be traced, and meet the original system-level requirements.
According to Hemmelgarn, integration between the two companies’ products is already working out very well for customers. Siemens PLM’s CAD products, for example, are brought into Mentor’s wire harness and cabling designs. Similar integrations are happening in PCBs and even in the integrated circuit space, he noted.
Integration isn’t taking place just on the conceptual level. An aircraft manufacturer in Europe, for example, has already seen value in the tight integration of the two companies’ products, claimed Hemmelgarn. By “bringing together Siemens’ CAD product with Mentor’s cable and harnessing capability,” the manufacturer did “3D routing throughout the aircraft,” he explained.
AI coming to the factory floor?
Notably, Siemens has a simulation model for everything — “every Functional Mock-up Interface (FMI),” said Rhines. FMI is a tool-independent standard to support both model exchange and co-simulation of dynamic models using a combination of xml-files and compiled C-code.
Siemens also runs factories all over the world. That range makes Siemens’ cloud-based, open IoT operating system — called MindSphere — a key enabler connecting products, plants, systems and machines, Rhines claimed. Presumably, a gateway built on MindSphere makes it possible to collect a large amount of data generated by IoT with advanced analytics.
“You can now use IoT to leverage and understand analytics — what’s happening to this manufacturing process,” said Hemmelgarn.
But what about machine learning? We asked Hemmelgarn how far Siemens and Mentor have advanced AI into the connected digital factory.
Obviously, AI works not just in voice and vision, Hemmelgarn said. “Any kind of data can be collected.” Such data as vibrations or breakage observed in production-line machines, for example, can be collected and analyzed. The data can be used to monitor the factory. But the big question, Hemmelgarn noted, is: “Are we learning?”
Although Siemens has a group of PhDs working on various AI algorithms, “we are still in the very early stage [of learning],” he noted.
For example, there is a sequencing tool for “pick and place” [inserting components such as registers and capacitors onto a PCB] to reduce the travel time of placing such parts. But to optimize the process, ideally, algorithms must be able to “work not just on one machine but on a mix of machines,” Hemmelgarn added. Further, they must work under varying environments at different levels of humidity and temperature.
While “learning” across machines and different tasks remains tough to do, machines are already used in automated wiring for automotive, for example. By understanding wire lengths and the number of connectors needed, along with a vehicle’s cost and weight requirements, a machine can be tasked to look at thousands of options to optimize and automate wiring inside a vehicle.
“That’s a level of complexity that humans can’t keep up with,” said Rhines. Using a process called “generative design,” Mentor’s tools are already providing “a full electrical design from logical schematics through physical/topological wiring generation, and onto fully detailed and costed harness drawings,” according to the company.
Generative design was also applied to a high-end Bugatti sports car. On one hand, the carmaker applied traditional fluid dynamics to design the upper half of the model. On the other, the vehicle vendor leveraged generative design to identify light-weight materials for the bottom of the car, making both stronger and lighter.
Meanwhile, parts and components are automatically classified to make a “shape search” possible. The parts’ cost and weight can be added to the index.
Hemmelgarn explained that automation can take care of “classification.” But can the machine learn to automate this process for customers, tailored for their needs? Hemmelgarn said, “We are only scratching the surface.”
Safety validation beyond ISO26262?
Siemens-Mentor already has a range of simulation tools and verification software to get ISO26262 functional safety done right for tier ones and car OEMs. But can these tools verify and validate the safety of automated vehicles that use machine learning?
The Achilles heel of machine learning, as it turns out, is that much of AI’s learning is done in a black box. Designers of automated vehicles have no idea how a machine has come to a certain conclusion, let alone what exactly it has learned from a large, unsupervised dataset. Unless AI can explain what it’s learned (“explainable AI”), a classical “V-model” validation model doesn’t much help to ensure the safety of AI-enabled automated vehicles.
Both Rhines and Hemmelgarn acknowledged that it is a big challenge.
Hemmelgarn, however, indicated that there might be a way around the dilemma. Chips can self-test through dynamic queries placed by system experts who write tests. This exercise can expose weird things happening inside chips.
Similarly, by applying a host of simulations to automated vehicles, if an automated vehicle is behaving strangely in certain situations, the system can alert car designers to “watch out for this or that,” noted Hemmlgarn. “By keeping hardware in the loop,” he explained, some AI pitfalls could be prevented.
— Junko Yoshida, Chief International Correspondent, EE Times