Semiconductors and Components

Power Efficiency to the People, ARM CEO Says

07 July 2015

Simon Segars, CEO of intellectual property licensor ARM Holdings plc (Cambridge, England) has issued a call for the industry to solve the power consumption challenge, which he says leads the list of barriers that must be cleared if the Internet of Things (IoT) is to fulfil its promise.

His comments suggest that even ARM, a company focused on power efficient processor cores, is feeling pressured from customers. Segars appears happy to push those market concerns into the research ecosystem.

Simon Segars, CEO of ARM.Simon Segars, CEO of ARM. ARM is a key player in such collaborative ecosystems and Segars issued his call from the speaker’s platform at the IMEC Technology Forum in Brussels in late June. His company ships in billions of chips each year and has focused on power consumption and power efficient computation since its formation in 1990. As a result, ARM has been able to surf the wave of mobile phones, smartphones and tablet computers.

Other processor manufacturers such as Intel and Advanced Micro Devices Inc. previously had tended to pursue the maximum performance for processors that could be sustained by air cooling in a personal computer. By contrast, ARM had a more power efficient architecture and was selected for use in battery-operated, mobile devices, which are sensitive to power consumption.

The gap between the x86 and ARM architectures has closed over the past five years, however, as Intel has worked to produce processors with improved power efficiency and as ARM has sought to address higher-performance markets such as servers.

Computing follows a simple model, Segars says, that includes a central processing unit, some memory and some I/O circuitry. At this level of abstraction, what has really changed is the I/O, which increasingly is done wirelessly, he says.

Lower Power, Better Packaging

The billions of end-point nodes in the IoT will present different computational and communication requirements. They will require a processor and software to be sure, but the main drivers will be cost and power consumption. There will be many and varied sensors, but also a need for power-efficient communications and computation that sleeps a lot. The cost must be in the range of cents per node rather than dollars per node, and all must be made secure with authentication and trust, Segars says.

For ARM, a supplier of digital processor IP, this means greater emphasis on ultra-low power. This often translates to ultra-low voltage of operation and on MEMS, analog circuits, radio frequency (RF) and embedded flash or an equivalent non-volatile memory technology.

As an example, Segars points to a subsystem that would link a Cortex-M3 processor core and ARM's recently acquired Bluetooth radio IP optimized for the 55nm ULP manufacturing process from foundry Taiwan Semiconductor Manufacturing Co (TSMC). The entire subsystem can operate at about 1 volt, he says (see ARM Offers Optimized IoT Subsystem).

For such an application, a move to smaller circuit dimensions is not as important as the move to lower power consumption. Segars says that a move from 1.2V to 1.0V increases battery life by 60%. However, as voltages drop and approach transistor-threshold voltages, the leakage current goes up and transistors become more sensitive to temperature and manufacturing variability.

Embedded Memory

Segars also addresses embedded memory as a second barrier that needs to be cleared. The advantage of non-volatile memory is that nodes can be switched off for large proportions of the time to save power but resume operations as soon as the node "wakes up."

The problem with the incumbent technology—embedded flash memory in use down to 40nm node—is that the memory cell takes up a relatively large die area, requires high voltages (which implies the use of a charge pump that also takes up die area), and is challenged in providing enough endurance cycling for some applications. The result is often costly and unreliable memory that does not scale easily.

IoT needs a non-volatile memory with low voltage operation that does not require extra steps for integration with CMOS logic, says Segars. A number of contender for non-volatile memory technologies exist, but they generally lack a track record of reliability. Even if they prove to be technically sound, they may take time to penetrate the industry. These technologies include the NRAM from Nantero Inc., which is based on resistance switching in a carbon nanotube matrix (see Carbon Nanotube Memory Enters Design Phase) and multiple examples of resistance switching in metal-oxide stacks and voltage induced metal-ion migration as a method of making and breaking conducting filaments with cross-points (see Crossbar Edges Closer to ReRAM Debut).

Such novel technologies are often beset with a chicken and egg problem; without clear market demand, EDA and IP companies are reluctant to invest resources to support novel technologies. Without that support, however, novel technologies struggle to gain design wins.

Packaging Technologies

Segars’ third request to the R&D engineers was for packaging technologies. With multiple die—MEMS die, CPU die, RF die—in a component, there needs to be a method of connecting them. He says that through-silicon vias (TSVs) are in their early stages and research to date has been performance-driven. Segars calls for an interconnect system that can meet a diverse range of applications that can be dense, low cost and high volume.

An IoT with tens of billions of devices will need a different network architecture, says Segars. That means there is going to be a "fog" of distributed computing up and down the network with as much computation done as near to the point of use as possible to minimize computational power consumption and the bandwidth of transmitting data. He adds, "How can I compress it? How can I encrypt it and how can I compute closer to the edge? That's what 5G is all about."

Networks will have to evolve and will require large heterogeneous processors that will demand 10nm manufacturing and more advanced geometries, he says.

"Computing is going to become a different thing,” Segars says. “It will be all around us but it also needs to become invisible so consumers won't need to know about it."

Questions or comments on this story? Contact

Related links and articles:

IHS Technology Semiconductors and Components Page

News articles:

ARM Offers Optimized IoT Subsystem

Carbon Nanotube Memory Enters Design Phase

Panasonic, IMEC Take Embedded ReRAM to 28nm

Crossbar Edges Closer to ReRAM Debut

Powered by CR4, the Engineering Community

Discussion – 0 comments

By posting a comment you confirm that you have read and accept our Posting Rules and Terms of Use.
Engineering Newsletter Signup
Get the Engineering360
Stay up to date on:
Features the top stories, latest news, charts, insights and more on the end-to-end electronics value chain.
Weekly Newsletter
Get news, research, and analysis
on the Electronics industry in your
inbox every week - for FREE
Sign up for our FREE eNewsletter
Find Free Electronics Datasheets