Embedded systems power down

Low-power sensors, microcontrollers, and communications circuits combine to acquire, process, and transmit data in applications ranging from environmental monitoring to health care.

Rick Nelson, Editor-in-Chief -- EDN, July 29, 2010

Those sensors in turn will produce a lot of data for processing. Perhaps that processing will take place in remote server farms, but it’s a good bet that at least some processing will occur within embedded microcontrollers near the sensors themselves. The resulting data then will need transmitting—often over a wireless link. The entire operation—sensing, processing, and communications of the results—must take place with extremely low power consumption, affording low heat dissipation and long battery life or, perhaps, elimination of the battery through energy-harvesting techniques (see sidebar “Algorithm measures battery’s state of charge”). Design-automation tools will be necessary to help bring sensor-based products to market in a cost-effective and timely manner. Since Conner’s May 27 cover story, participants at two events have addressed these issues. The IMEC Technology Forum took place in June in Eindhoven, the Netherlands, and Leuven, Belgium; the DAC (Design Automation Conference), took place during the same month in Anaheim, CA.

A sustainable world

Speaking at the IMEC Technology Forum, IMEC’s president and chief executive officer, Luc Van den hove, painted a picture of a world that will look significantly different in 2025. Innovations will span high and low tech, he suggested, with radiant tiles dimming automatically to control electric costs, with rooftop grass providing cooling and turning greenhouse gases into oxygen. Not surprisingly, IMEC is focusing on the high-tech solutions. “Working toward a sustainable world is one of the most important issues our society is facing,” Van den hove said, explaining that energy use has increased 35% over 10 years. “We have to adapt our lifestyle to use less energy,” he said, adding that the smart grid will be a crucial technology in that effort. IMEC will contribute to smart-grid and other green-energy implementations with, for example, its research into gallium-nitride power devices and materials for super capacitors, batteries, photovoltaics, and fuel cells.

IMEC is placing much effort into health care. Emerging technologies will enable pregnant women to receive weekly checkups at home, and smart labeling will enable them to choose the right foods. Researchers at the Holst Centre, a research affiliate of IMEC, apply technology to meet societal challenges and needs. They elaborated on relevant health-care efforts and the sensor technologies they are pursuing that will help bring Van den hove’s vision to reality (Reference 2).

Jo De Boeck, senior vice president of smart systems and energy technology at IMEC and a director at the Holst Centre, provided an overview of the organization’s efforts to reduce energy consumption and promote electronic-health initiatives. The technologies under development include innovative sensors and actuators, ultra-low-power DSPs, low-power analog circuitry, ultralow-power wireless technology, and micropower-generation technologies.

Health-care opportunities

Health care offers particular opportunities as aging and chronically ill people worldwide tax health-care efforts. Worldwide, 1 billion people are overweight, 1 billion people are 60 years of age or older, and 1 billion people are chronically ill. The Holst Centre’s Human++ initiative provides at-risk populations with preventive, predictive, personalized, and participatory health-care technology. The ultimate goal is a closed-loop system involving symptoms, testing and monitoring, diagnosis, and treatment. In support of the effort, the Holst Centre is developing ultra-low-power technology that eliminates or reduces the need for battery replacement, as well as unobtrusive sensors and other devices that are implantable or comfortable, wearable, and stretchable. The Centre is working on sensors that can analyze sweat, breath, saliva, and a patient’s environment.

Julien Penders, program manager for body-area networks in the Human++ program, provided some examples of the Centre’s efforts by describing IMEC’s ECG (electrocardiogram) necklace. The wearable device provides continual monitoring for a week and offers a body-area-networking interface to an Android phone, making data available over the Internet; the device can send warning e-mails or text messages. One application for ECG monitoring is epileptic-seizure detection, said Penders, adding that 50 million people worldwide suffer from the condition, which costs $15.5 billion in direct and indirect costs each year in the United States alone. IMEC’s low-power ECG system includes an algorithm that recognizes heart-rate patterns associated with seizures and notifies health-care providers.

Other efforts center on arousal and stress monitoring, based on sensor technology that monitors the autonomic nervous system. Studies have monitored the psychological state of day traders, who may be on the verge of making irrational deals, and chess players. Relevant multimodal sensing technology in such applications monitors such factors as skin conductance and ECG, with work under way on adding EEG (electroencephalogram) and the sensing of chemicals such as cortisol, a biomarker for stress.

Body-area networking

To be effective in wearable and implantable applications, sensors and the associated interfacing technology must be low-power, said Bert Gyselinckx, general manager for the Holst Centre/IMEC and program director for Human++. To that end, the Holst Centre has developed a body-area-networking toolbox employing nanowires, nanoparticles, clamped beams, capacitive and oscillator readouts, processing, energy harvesters, and wireless transceivers. A power-estimation tool describes the power performance of combinations of such tools. A radio can account for about half of the power such a system uses, so IMEC and the Holst Centre have been focusing on wireless technology with the goal of outperforming standard wireless technologies such as Zigbee and Bluetooth with respect to power consumption. Gyselinckx described an IMEC-developed event-driven radio that operates on less than 0.1 mW and a 2.4-GHz body-area-network transceiver that offers data rates of 64 to 1024 kbps. It employs OOK (on/off-keying) signaling and a superregenerative-receiver architecture. It draws 1 mW as a receiver, 2.5 mW as a transmitter at 0 dBm, and only 0.9 mW transmitting at −10 dBm. Its range is 3 to 5m.


Crego-Calama also described a lowcost manufacturing approach, in which ink-jet-printing technology can deposit on the MEMS structures the polymers sensitive to chemical vapors. The e-nose can monitor air quality; identify pathogens; monitor food ripeness or spoilage; monitor physiological conditions from breath analysis, for example; and detect biological or chemical weapons.

Processing the results

Whatever types of sensors you are deploying, you’ll need to process the results—for data logging, to generate alarm signals, or perhaps to remind patients to take their medicine. The approach that can most quickly get you to market is to employ off-the-shelf low-power microcontrollers, such as the Texas Instruments TMS320C2000 Piccolo real-time-control microcontrollers. Such an approach also helps you take advantage of off-the-shelf software tools that can help you develop your algorithms and port them to the target controller.


Previewing the announcement at DAC, Ken Karnofsky, senior strategist for signal-processing applications at The MathWorks, said that creating energy-efficient designs is quickly becoming the most important part of embedded development. “Model-based design helps embedded-system developers reach the market faster with innovative products that harness the lowcost, high-performance, and controloriented peripherals of Piccolo microcontrollers,” he explained.

The MathWorks/TI collaboration enables engineers to design and simulate systems, generate code, and verify algorithms running on TI Piccolo processors using The MathWorks’ Embedded IDE (integrated-development-environment) Link tool. The MathWorks’ code-generation products enable engineers to integrate peripheral devices and real-time operating systems with algorithms they created using Simulink models, Stateflow charts, and embedded Matlab, all without writing low-level drivers and runtime code. The Math-Works’ Fixed-Point Advisor tool helps target floating-point implementations to a fixed-point processor. After deploying the resulting executable on a Piccolo microcontroller or another C2000 microcontroller from TI, engineers can perform processor-in-the-loop testing. To speed execution, engineers can replace parts of the standard ANSI C code with processor-optimized functions.

The 32-bit Piccolo family offers a range of performance levels, various flash options, analog integration, and control-oriented-peripheral options to meet the varying demands of cost-sensitive, real-time control applications. Design engineers can now execute their Matlab and Simulink algorithm code across all F2802x/F2803x Piccolo devices for rapid prototyping and production deployment of embedded systems. Model-based design with production-code generation creates a direct connection between the development environment and the implementation platform, helping engineers to identify and fix design problems at the system level and easily generate efficient C2000-specific code.

Microtasking=low power

Off-the-shelf, general-purpose components don’t, however, give you optimal power savings, which requires a more customized approach. Speaking at DAC, Muhammad Adeel Pasha, a professor at the University of Rennes and researcher at IRISA (Institut de Recherche en Informatique et Systèmes Aléatoires)/INRIA (Institut National de Recherche en Infomatique et en Automatique), described how he and his colleagues are developing a design flow for the generation of ultra-lowpower wireless-sensor network-node architectures employing “microtasking” (Reference 5). Microtasks become active on an event-driven basis. Effective implementation of a microtask requires hardware specialization to reduce dynamic power and power gating to reduce static power. Pasha described a complete design flow for microtask implementation from C down to synthesizable VHDL, adding that initial estimates show power savings of one to two orders of magnitude better than that of microcontroller-based approaches.

Several presentations at DAC in addition to Pasha’s focused on low-power techniques. For example, Adam Cabe, a professor at the University of Virginia, described work he and his colleagues have performed to reduce SRAM power consumption in standby mode—a critical issue because SRAM caches can dominate the total area of some chips (Reference 6). To minimize the idle currents SRAM banks draw when inactive, the University of Virginia researchers employ an implicit voltagereduction scheme that requires no on-chip dc/dc converters. The approach essentially stacks inactive SRAM blocks in series in a way that maintains the required data-retention voltage on each SRAM block and reduces leakage power by 93%, according to simulations in 65-nm technology.

In yet another presentation focused on low power, Weixun Wang, a professor at the University of Florida, described how he and a colleague, Prabhat Mishra, implemented an intratask pre-emptive dynamic-voltage-scaling scheme that can cut energy consumption by as much as 24% in contrast to savings through intertask voltage-scaling techniques (Reference 7).

Academic initiatives such as those that Pasha, Cabe, and Wang described indicate what is possible, but it takes additional effort to bring products to market. Representatives of several companies commented on the commercial aspects of low-power design in a DAC panel discussion that Rob Aitken, an R&D fellow at ARM, moderated (Reference 8).

S Balajee, an engineer at Texas Instruments India, described what’s necessary as we move to an always-on and always-connected mobile world, including standards to capture power intent; power-profiling tools; design-verification, synthesis, and place-and-route flows that address multiple voltage domains; and sign-off tools for validating power-efficiency intent. Toshiyuki Saito, an engineer at Renesas Electronics Corp, cited the importance of “3S”: switching activity, size of a chip, and supply voltage. He noted the effectiveness of DVFS (dynamic voltage and frequency scaling) as a low-power design method in which components outside an SOC (system on chip) usually control frequencies and voltages. Saito also mentioned A-DVFS (adaptive DVFS), in which components within the SOC adaptively control supply voltages, and he called on EDA vendors to take on a key role in supporting efficient design of low-power systems. Venugopal Puvvada of Qualcomm India emphasized that an effective power-management scheme for an IC requires a system-level view with accurate power-simulation capability.

EDA vendors weigh in

Panelist Koorosh Nazifi, an engineer at Cadence Design Systems Inc, said that EDA vendors are providing automated advanced low-power-design techniques that were once the domain of sophisticated designers with centralized in-house CAD systems. He conceded that this initiative requires more work, especially with regard to mixed-signal SOCs and system-level architecture exploration.

“Energy consumption is a critical design constraint,” said panelist Allan Gibbons, principal engineer at Synopsys. “Dark silicon is a real and present challenge. We must be smarter about how we spend our energy budget, and don’t push for gigahertz with no regard for energy.” You must minimize energy cost in watts per MIPS (millions of instructions per second) when devices are in active mode and avoid wasting energy when devices are doing no useful work in standby mode. “Power consumption is replacing performance as a key figure of merit at [processes of] 32 nm and below,” he added. Achieving an optimal approach for energy efficiency will require power-aware exploration with virtual prototyping, allowing designers to concurrently optimize all components of a design.


Van der Perre questioned how we can keep manygigabit-per-second communications going. The answer, she said, lies in compact, lowcost, low-power cognitive reconfigurable multimode radios that include spectrum-sensing features to support dynamic use of spectrum. The IMEC approach centers on the Scaldio scalable-radio analog front-end and the COBRA (cognitive-baseband-radio-architecture) digital baseband device (Reference 9). The COBRA processor supports 1-Gbps concurrent streams, multithreading with SIMD (single-instruction/multiple-data) capabilities, a FlexFEC (flexible forward-error-correction) processor, and a front end that supports flexible filtering and spectrum sensing. A COBRA template supports fine-tuning to end-user requirements. According to Van der Perre, radio algorithms and architectures are ready to support 4G (fourth-generation) requirements, with sufficient flexibility in resampling and programming filters to meet emerging cognitive-radio, sensing, and multiband-reception requirements.

System-level needs

Universities, companies, and research organizations will continue to pursue low-power sensors, processors, and radios. All could use a boost from EDA vendors. Iqbal Arshad, corporate vice president of innovation products at Motorola Mobile Devices, summed it up during a June 17 DAC keynote address at which he explained how his team developed the Motorola Droid, describing the synthesis of new hardware that tightly couples with new software into a global multiband product. The team did not have all the automated design tools it might have wished for as it pursued its development schedule at breakneck speed, Arshad said, relying instead on reference designs. Building a board is easy, he said, but integrating the entire system was difficult, involving interrelated electrical, mechanical, and other design considerations.

Power-efficient design proved to be particularly challenging. According to Arshad, SOC suppliers tend to think power management ends with their chips, but they are not the key considerations for end users. “There is no tool today to do effective power management for the end product,” he said, pointing a clear path to where DAC exhibitors might turn their attentions over the coming year.

You can reach Editor-in-Chief Rick Nelson at richard.nelson@cancom.com.

---

References

Conner, Margery, “Sensors power the ‘Internet of Things,’” EDN, May 27, 2010, pg 32. Nelson, Rick, “Health care is focus of Holst Centre research efforts,” EDN, June 8, 2010. Nelson, Rick, “Low-power e-nose checks air quality,” EDN, July 15, 2010, pg 14. Nelson, Rick, “MathWorks adds model-based design for TI’s C2000 Piccolo microcontrollers,” EDN, June 23, 2010. Pasha, M, et al, “A Complete Design-Flow for the Generation of Ultra Low-Power WSN Node Architectures Based on Micro-Tasking,” DAC Proceedings 2010. Cabe, A, et al, “Stacking SRAM Banks for Ultra-Low-Power Standby Mode Operation,” DAC Proceedings 2010. Wang, W, and P Mishra, “PreDVS: Preemptive Dynamic Voltage Scaling for Real-Time Systems Using Approximation Scheme,” DAC Proceedings 2010. NS, Nagaraj, “What’s Cool for the Future of Ultra Low Power Designs?” DAC Proceedings 2010. Craninckx, J, and P Wambacq, “Reconfigurable single-chip radios,” EDN, May 27, 2010, pg 29.

---

For More Information
ARM	Holst Centre	Qualcomm	The MathWorks
Cadence Design Systems Inc	INRIA	Renesas Electronics Corp	University of Florida
DAC	IRISA	Synopsys	University of Rennes
IMEC	Motorola	Texas Instruments Inc	University of Virginia