Analog’s Rising Status

Uptick in demand low power and power-sensitive designs brings new challenges and opportunities.

popularity

As more sensors and actuators are added into electronic devices, pressure is growing to more seamlessly move data seamlessly back and forth between analog and digital circuitry.

Analog and digital always have fit rather uncomfortably together, and that discomfort has grown as SoCs are built using smaller feature sizes. While digital transistors can continue to scale to well below 28nm—there is debate now about just how far the digital roadmap will continue—analog is moving at its own pace. In fact, in many cases it isn’t moving at all.

But if they don’t move at the same pace, at least they have to talk better. That helps explain why at 28nm and 16/14nm, standard “analog” IP includes a fair amount of digital content. And as the IoT pushes up demand for analog content, adding sensors to connect the physical world with electronic devices, the need for even greater interoperability and communication between these two worlds will continue to grow.

“The demand for analog silicon has always existed in the embedded space, but the advent of the Internet of Things (IoT) is increasing the demand for connected mixed-signal content,” said Ian Smythe, director of marketing programs for ARM’s CPU Group. “Sensors are making big changes to deeply embedded systems in a range of smart technology, in the forms of smart devices, smart homes, smart buildings and smart cities.”

To enable the information to be processed and transferred efficiently, analog data needs to be processed more efficiently in those embedded systems. That includes everything from low-power microcontrollers that are being deployed in edge devices, more complex and powerful processors used in home and building gateways, and the so-called fog servers sitting between those gateways and the cloud.

One reason why communication needs to improve is that most of the IoT’s interesting capabilities depend on the ability to combine data from multiple sensors.

“The sensors are often MEMS or silicon photonics or whatever it is that matches up with the physics you are trying to sense,” said Jeff Miller, product marketing manager at Mentor Graphics. “The analog is the piece that creates that interface between the sensor and the digital system that’s going to do things like sensor fusion to bring in data from multiple sensors, combine that with useful information, and then pass it along.”

Connecting analog and digital
Most of this fits under the general headings of IoT or connected devices or smart cars, but the basic challenges are the same. A smart car has loads of sensors. So does the industrial IoT, and so does infrastructure monitoring for a bridge to tell whether it is above its design load. There is also a lot of activity in the medical area, with a big next wave of connected sensors coming for medical in the area of consumer-level health monitoring systems.

This rise of smart analog is resulting in processing capability continuing to be added to analog components, Smythe observed. “Today, many sensor devices have processors inside, often used to automatically calibrate analog components to improve accuracy and performance. Not only is this essential at the manufacturing stage, but continual calibration also boosts performance throughout the life of a product and can even extend it.”

Take a simple street light sensor, for example. If it gets covered with dirt, it can turn the light on too early. Moreover, the light lens itself becomes obscured and the effective light becomes dimmer over time. With better communication between digital and analog, lights can be automatically turned down when the lens is clear, and turned up as the lens becomes more opaque.

The automotive industry also has new requirements on analog/mixed signal designs, Smythe noted. For example, a range of safety features require various forms of sensors to provide information about the vehicle and the environment. “As cars become more autonomous, the demands on sensors and systems that process sensor data grow ever stronger. For these applications, functional safety and the reliability of the design are critical.”

Within these vehicles, the inputs to an ADAS (cameras, sensors, radar, laser, sonar) and the outputs (navigation, breaking, airbag trigger, collision avoidance) are all analog/mixed-signal designs.

“The key verification considerations are robustness and reliability to meet the high standards of ISO 26262,” said Geoffrey Ying, director of marketing at Synopsys. “Such systems must work as expected in various scenarios, which requires the designs to be thoroughly simulated and regressed against PVT, as well as any potential fault conditions and recovery.”

Analog in the cloud
The growing emphasis on mixed signal doesn’t stop there, either. The push for unlimited communication bandwidth in data centers includes analog components, as well.

“To support such demand, as an example, new optical driver ICs with 4-level Pulse Amplitude Modulation are now available to support up to 400 Gbps Ethernet,” Ying said. “These chips have a high level of analog and digital integration, are operating at extremely high data rate, and yet must keep power low to conserve energy consumption. To verify such chips, designers need an accurate mixed-signal verification solution with high simulation throughput and high capacity, as well as the ability to accurate signal integrity analysis of the architecture.”

What’s tricky about all of these applications is that all of the decision making and all of the smarts are in the digital domain, said Mick Tegethoff, director, AMS product marketing at Mentor Graphics. “In the case of a car, it’s a microcontroller. In the case of IoT devices, it’s a mini-mini-mini-microcontroller, which is digital.”

As such, there is a lot of activity around the fast and accurate data conversion from analog to digital, and the other way around.

He also noted that in many designs, there is more wireless technology. “In the case of IoT, it’s all wireless. And again in cars, there’s more wireless. Whenever you go communicate via WiFi or anything, you get into the analog RF domain. Traditionally, ICs for cellular and WiFi have always had analog RF content when you move away from the digital to interconnect with each other via the cellular infrastructure or the wireless infrastructure, and that continues to grow. Demand on those devices continues, and for technologies like 5G, microwave is required — aka analog on steroids. It is very complex.”

Back in the spotlight
For Mladen Nizic, engineering director for mixed-signal solutions at Cadence, this is a golden age for analog/mixed-signal again, with automotive and IoT specifically driving much demand for integrating analog at every level to really make systems work properly. But while the trend to move toward the digital world as much as possible has not really changed, there are myriad new challenges.

“On one side there is a trend to digitize as many analog functions as possible, but at the same time, there are a number of touching points,” Nizic said. “These systems’ interaction with the real world has increased, so something that before had one or two sensors now has a dozen or more. This requires a higher level of integration. If you take IoT edge node devices that collect the data, depending on the functionality, it might have to do some amount of the processing. How much processing needs to be done really depends on the amount of information that’s being gathered, along with bandwidth efficiency of the interface to the gates and server, and how much information it makes sense to transmit. Some amount of processing has to be moved to the edge node, and in some cases it demands more advanced process node and compute power there. In other cases, it could be quite simple but it has to be very power-sensitive. These IoT edge nodes suddenly become quite diverse and versatile systems that have to work reliably over many years on minimum power. Analog became a key element in making sure these really work properly, because it has to work reliably combined with safety standards, minimized power consumption as well. That is driving a lot in the IoT.”

As always, the verification is quite a challenge. “The digital methodology has been well established, but now what makes it even more challenging is that the analog cannot be considered in isolation,” he said. “It has to be considered as part of that system in the verification process. And, it has to start early. It has to start almost concurrently with digital. It may be that the analog circuitry is not even designed yet, so that’s why what used to be the standard practice in digital verification is becoming a strong requirement increasingly for analog/mixed-signal, as well. This means good, strong verification planning from the beginning, developing the right verification and simulation strategy. Here too, analog modeling is taking off very rapidly.”

Then when it comes to physical implementation, there must also be a tightly integrated solution for the physical design, starting from floor planning, through the layout, and then sign-off. “This is an area where we see a strong uptick, because in the signing off of performance as well as timing, for example, electromigration and IR drop is becoming critical and has to span across both analog and digital,” Nizic said. “It cannot be done separately.”

The impact on designers
There are other considerations for the engineering team, as well. When digital engineers need to interface with some kind of physical system, there are often constraints that are not obvious.

“A lot of times, this analog work is being done at process nodes that are perhaps a little surprising to some people,” Miller said. “We’re seeing a lot of work happening on 180nm, 90nm, and these kinds of process nodes rather than 22nm. This might seem a little counterintuitive, but it turns out that doing analog at those very small process nodes imposes a bunch of challenges that require a lot of extra work to make something function at those process nodes, as well as a lot of extra expense. And it turns out that you don’t get any of the benefits, so the only reason that anyone would go to these finFET nodes is really if they have to integrate with really large scale digital and you can’t do it off-die. There are certain cases where that’s necessary—a PLL on an application processor. But for the most part, wherever it’s not necessary, people are focusing on the nodes that really do analog well. Those are the larger feature sizes, and you get a cost savings advantage.”

For straight analog performance, the only engineering teams going to the extreme nodes are those with extreme power consumption requirements, but some of the larger process nodes still do a very nice job of that, he added.

Underlying forces
“We started 20 years ago talking about these things called SoCs,” said Drew Wingard, CTO of Sonics. “And the name literally tried to imply shrinking the entire electronic function of the system to put it on a die. Obviously for the vast majority of applications, we never actually got there, but integration continued. It’s not like we ever stopped integrating more stuff on the chips. We just found other ways of doing it, or we found more functions. We could spend that silicon area on adding more functions in a digital domain to the big digital chip, and continued to keep some of the mixed signal stuff off on a different chip. Since it appears that in a lot of application areas it’s no longer obvious what to do with more transistors for a standard digital logic perspective, we begin to look much harder at things like cost. The end application is enabled by having a chip that has a really long battery life and costs a dollar. As soon as you do that, it’s, ‘Oh my gosh, I’d better integrate because every extra die with a package costs this amount of money, and I have to look at the total bill of materials for the system,’ so the pressure for integration gets to be quite a lot higher.”

That creates demand by chipmakers to pull in mixed-signal content that would have been there, but would have been on a separate die, and probably in a separate package, Wingard said. “Even if we’re in the place that integrating some of this mixed signal process technology still isn’t economically wise, we’ll integrate it using some form of stacking. We’ll do something that achieves the cost goals by bringing them closer where the connection between them doesn’t look like a bond wire to a standard board level interconnect. It makes them more of our design. So you start to think about what is optimized to match with what you’ve got. It becomes part of a larger multi-die design process.”

Conclusion
Demand for analog is growing across all applications. The performance and power requirements — whether it is high power in one end, and low power in the other — continues to grow. The technology, whether it is BCD for automotive, or 7nm for the SoCs introduce new effects that need to be dealt with in the transistors, and in the simulation.

“Analog is still very much an art form as opposed to a push button synthesis flow on the digital side,” said Tegethoff. “It’s very much designer-dependent, and about designer creativity skill and knowledge experience.”

So no matter how much analog and digital need to get along, there will always be differences. But increasingly, two engineering teams will be required to mask those differences and make communication between these two worlds seamless.

Related Stories
Can Analog And Digital Get Along Better?
Combining both in a mixed-signal design brings challenges in a different realm. Expertise is the key to success.
Rethinking The Sensor
As data gathering becomes more pervasive, what else can be done with this technology?
Culture Clash In Analog
Different design perspectives begin creeping into analog world as systems engineers enter the market with big-picture ideas and training.
Mixed-Signal/Low-Power Design
Part 2: Getting analog and digital design teams to work together; the chip company executives challenge the EDA vendors.
Mixed-Signal/Low-Power Design
Part 1: Adding ultra-low-power requirements to a device design is complicating the traditional process of mixed-signal IC design.



  • Kev

    A major problem with the design flow is that it is dominated by the digital guys that think in 1s & 0s. Having worked on the Verilog-AMS standard from the get-go, I can say there has never been any great enthusiasm from the EDA companies to make it work properly – Cadence screwed up the original implementation deliberately, and Synopsys never did a full implementation. As such, anyone expecting working tools for AMS SoCs in the near future is probably SOL.

    On the upside neural networks look a lot like analog circuits and people are building ASICs for computing those fast. However, you do need to know how to write the behavioral analog models 😉

    – actually there are AI techniques for generating the models, so it isn’t nearly as hard as people think…