Millimeter-scale nearly perpetual sensor system
By all accounts, wireless sensing is a growing field, with new technologies enabling sensors to operate in remote locations and under challenging conditions. These advances have created a deluge of emerging applications, of which many are highly volume constrained. For example, doctors can implant millimeter-scale medical devices into the eye through minimally invasive surgery. These sensors can measure eye pressure to track the progression of glaucoma and other ocular diseases.
We have proposed a sensor that will deliver continuous pressure and temperature measurements (which represent just two of many possible sensing modalities). The measurements are digitized using capacitance- and time-to-digital converters, and delivered to an on-sensor processor that performs signal processing and logs measurement results into memory. The system is powered by solar cells and a thin-film lithium battery.
Although miniature sensor nodes have been proposed in the past, they have all suffered from short device lifetimes.1,2 The root cause of this is that it is impossible to store an appreciable amount of energy on such a tiny sensor node. Even with the most advanced battery technology, a millimeter-scale system using commercial circuit techniques would only operate for minutes before draining its power supply. To extend lifetime from minutes to decades, our system uses solar-energy harvesting to continuously recharge a battery (see Figure 1).3 Ultralow-power operation enables us to bridge year-long gaps when no solar energy is available.
Our system harvests energy under bright indoor- to sunny outdoor-lighting conditions. Solar energy is harvested on two series-connected silicon photovoltaic diodes. The solar cells have an output of 20nA–2μA at 1V. However, most common battery chemistries have higher voltages. We have chosen a 3.6V 12μAh (ampere hour) lithium battery supplied by Cymbet™ for the system, since lithium chemistries have high energy density. To recharge the battery, we use a switched-capacitor network (SCN) to pump up the voltage from the solar cells. Since the solar current is low because of the small photovoltaic area, this charge pump must be heavily optimized to prevent consumption of more energy than it converts.
While solar-energy harvesting generates a nearly infinite power source, its availability is sporadic. Our system must be able to operate, or at least survive, during extended periods when there is no light. In the glaucoma-monitoring application, these zero-light conditions occur when the patient closes his eyes and goes to sleep. Our system uses ultralow-power operation to take sensor measurements without quickly depleting the battery.
For ocular-pressure sensing and many other applications, readings taken every 10 or 15 minutes represent a ‘continuous’ measurement. This frequency is very slow on the timescale of the sensing and processing circuits used in the system. Thus, for the majority of time the system is idle, and most of its energy is consumed in standby mode. To reduce this dominant power, we use aggressive power gating.4 The power gates are very-low-leakage metal-oxide-semiconductor field-effect-transistor devices that disconnect the circuits from their power supplies.
Power gating reduces leakage power to femtowatts (1fW = 10−15W), but cannot be used on circuits that need power during standby. The latter include memory, a sleep timer, and a power-management unit, which rely on other power-reduction methods. Memory on chip must remain powered to retain logged data during standby mode. It is implemented as static random-access memory with low-leakage, thick-oxide, high-threshold-voltage transistors, resulting in a power of 3.3fW per bit. The sleep timer, which we have implemented as a 63pW leakage-based ring oscillator, controls the period between sensor measurements. Finally, the power-management unit must convert power between the solar cells, battery, and circuits. All of these transformations are performed with a single SCN that has an adaptive switching frequency to accommodate a wide variety of lighting conditions and load requirements.
During active mode, sensor-node tasks are performed with an advanced reduced-instruction-set computer machine (commonly known as ‘ARM’) Cortex™-M3 processor. The commercial instruction set provides a platform that is easy to program for end users. The active-mode power of the Cortex-M3 is reduced tenfold using voltage scaling.5
Our nearly perpetual system provides an important step towards millimeter-scale sensing, using energy harvesting and ultralow-power operation to solve many power-related challenges. However, many barriers still remain before a commercial product can be realized. We will add more sensing modalities to the system by using micro-electromechanical system sensors. We will also explore wireless communication, which is vital for communicating data to the end user when the system is not easily accessible, but creates new peak-power and transmission-distance challenges. Our power solution, along with these future circuit advances, will enable development of millimeter-scale sensors that provide a wealth of useful data in a wide range of sensing applications.
Gregory Chen received his BS and MS degrees in electrical engineering from the University of Michigan in 2006 and 2009, respectively. He is currently pursuing a PhD in electrical engineering. His research interests include power management for low-power systems, energy harvesting, and robust static random-access memory design.
