Get email delivery of the Cadence blog featured here
The Internet of Things (IoT) can be many things to many people, but to Steve Schulz it’s an enormous opportunity, wrapped in complexity, covered in the need for a new way of thinking about design.
Schulz (pictured right), the former CEO of Si2, came to Cadence Jan. 28 to share a vision for how engineering teams need to approach IoT and to re-think how they design for it.
It is, in an historical sense, a happy confluence of technological evolution (Moore’s Law plus decades of optimization in power, performance, and area) and market opportunities.
“IoT is about connecting things with information created by things to help decision-making by things... to benefit humans,” Schulz told the audience. “Ultimately, it’s the automated reconfiguration of all the things in society around us.”
To really exploit the potential, engineers need to embrace a system architecture view of the world, even when designing silicon. This is because IoT is about data—where data is captured, processed, and shunted and how it is most optimally computed up and down the line, from device to cloud.
“Data can be processed at different layers and there is a cost function to data. Where is the most efficient place to do the computing?” Schulz said. “The way you architect your communication of devices to layers of processing all the way up to the cloud--big data--that is going to affect what processing is done where, what power consumption occurs, how you optimize for what kind of decision-making is being done."
Additionally, the characteristics of the marketplace (a seemingly infinite number of application possibilities, huge competition, high product and standards churn, and the need for risk mitigation) cast some doubt on the “old” system-on-chip (SoC) way as the design path, he noted.
“Since the 1990s, we've assumed that if you're talking about integration, you're talking about ‘SoC.’ But now scaling is approaching economic limits for many markets. Integration must continue, but for IoT there are additional diversities there. So … those new integration requirements may force some changes to our design methodology."
These “diversities” include:
Here, Schulz sees the SoC era morphing, via 3D architectures, to the “system-on-stack” (SoS) era.
In this era, Schulz described an “early estimation” stack flow, if you will, that begins with product requirements and is followed by:
Iteration of this flow then follows, “and you get to something you can live with,” Schulz said.
In addition, contemporary design gaps must be closed or shrunk. Chief among them is low-power design, where Schulz said “we've exhausted a lot of low-hanging fruit” when it comes to optimization.
One way to push ahead on this front would be with the Multi-Level Power Model (MLPM).
The idea—which the Low Power Coalition has been working on for four years—is to have one single, consistent, and scalable model for all power simulation. This includes mapping system use cases that include software to the hardware modes and the TLM states. From there, engineers could drive power intent formats with the power intent and stimulus and perform ESL/TLM simulation, he said.
MLPM is a single power model with multiple external interfaces that allows it to support any abstraction level (cell, RTL, TLM), Schulz said.
“There's no reason it can't represent system-level interfaces as well,” he added.
The semantics for the TLM abstraction level for an MLPM standard emerged at DAC 2014, and proof of concept may emerge by this year’s DAC, paving the way for “more aggressive activity,” he added.
Schulz described other gaps that Si2 is working to close. With 2.5/3D IC flows, a technical advisory board is hard at work creating technical standards to address things like IP integration and electrical, physical, thermal, and mechanical considerations.
Work in silicon photonics standards in OpenAccess is underway with a joint effort by Si2 and the European Plat4m project, with an eye toward greatly improving bandwidth and efficiency for servers in the cloud. Schulz described this as a research project in its early stages.
Lastly, Schulz said work is continuing on emerging PDK needs for IoT design because more mature process nodes are attractive for such designs. Older PDKs will need to be refreshed (and refreshed more frequently) for low-power and ultra-low power considerations, he noted.
He noted that packaging, thermal, and 2.5D/3D data in PDKs might be leveraged to create a new “assembly design kit” standard.
—Si2 Conference: 25 Years of EDA Standards, and a New Role in SPICE Modeling
—Si2’s Steve Schulz: “Setting the Standards for EDA360”