Shrinking radios for millimeter-scale computers

Nearly invisible millimeter-scale systems could enable ubiquitous computing, and are the next class of computers predicted by Bell’s Law to fuel growth in the semiconductor industry.  Bell’s Law is a corollary to Moore's Law, which predicts a new class of smaller, cheaper computers about every decade. With each new class, the volume shrinks by two orders of magnitude and the number of systems per person increases. The law has held from 1960s' mainframes through the '80s' personal computers, the '90s' notebooks and the new millennium's smart phones.

Tiny, battery-operated wireless systems are the future of monitoring our bodies, environment, and buildings.  There are several challenges to realizing complete mm-scale systems, in particular in the design of the wireless interface.  Typical wireless radios have high peak power consumption, which cannot be sustained by micro batteries with peak currents of 10μA. They also use large off-chip antennas and precision crystal frequency references, which do not scale to volumes below 1mm³, or integrate with CMOS processes. In order to realize true mm-scale wireless systems, we are developing fully-integrated 1mm³ radios and antennas with strict limits on peak-power and energy/bit, and that eliminate the need for an external crystal reference.

Intraocular Pressure Monitor

Glaucoma is the leading cause of blindness in the world, affecting 67 million people worldwide.  It is caused by the elevated intraocular pressure (IOP) which damages the optic nerve. A mm³ sensor node inside the anterior chamber can be very helpful in monitoring the IOP. We have recently demonstrated a sensor node that is able to wake up upon reception of an external wakeup signal (downlink) and transmit recorded data back to the external unit (uplink). An asymmetric near-field (inductively coupled) wireless link is a suitable choice due to highly constrained area and power budget on the implant side.

Figure 1: Die micrograph of the wireless radio for the intraocular pressure (IOP) sensor.

The uplink consists of an oscillator with dual resonator tank. The dual-resonator architecture enables the two FSK tones to be far apart and thus relaxes the constraints on the phase noise of the oscillator. Obviously, bigger coils with multiple turns are desired to achieve higher transmission range. In order to provide enough negative resistance to oscillate a tank with these inductors and achieve cm-range transmit distance, tens of milliamps of power should be supplied from the power source.  Since the thin film battery peak current is only tens of micro amps, the transmitter operates off a total of 1.6nF of decoupling capacitors spread all across the chip. Upon transmission of every bit, the oscillator is turned off to give the capacitors enough time to recharge. This way, we overcame the peak current issue by trading off bitrate, which is not limiting in our system.

On the downlink side, the same dual-resonator tank is used to couple to the external unit. The voltages across the two tanks are rectified and compared against each other using a comparator with digitally-tunable offset. This approach eliminates the need for power-hungry voltage references and provides immunity to common-mode input fluctuations.

Antenna-Referenced Fully-Integrated Radios

In order to eliminate the bulky off-chip component and reduce the size of the wireless system, we recently demonstrated a fully integrated radio based on an antenna-referenced concept. The main idea is using an on-chip patch antenna as both the radiator and a frequency reference. The patch antenna operates in the 60GHz band, with dimensions of 1.2mm x 1.6mm. The frequency reference is the resonance frequency of the patch antenna, which is mainly determined by the width, length and height of the patch. The accuracy is robust over process variation because the metal variations in scaled CMOS is relatively small compared to the size of the antenna. Based on the measurement results of twenty replica antennas, a standard deviation of 1100 ppm on antenna resonance frequency variation is obtained. This tolerance is adequate for FCC compliance, and for non-coherent energy detection communication commonly used in tiny, battery-operated wireless sensor systems.

Figure 2: Die micrograph of the fully integrated 60GHz patch antenna with antenna-referenced frequency-locked loop

With the antenna reference, we built a frequency-locked loop to track the resonance of the patch antenna. By monitoring the difference in standing wave voltage magnitude observed at two tap points along the edge of the patch antenna that are equally spaced away from the length center, we can trace out a monotonic curve that crosses zero only at the resonant frequency of the antenna.  By using this as the frequency detector, we operate the antenna in a closed-loop controller to tune an oscillator to the appropriate frequency.

To further reduce the size of a complete mm³ system including processor, memory, and sensors, we implemented the patch antenna with a ground plane in metal 4.  This provides area for other circuitry and routing beneath the antenna that is shielded by the ground plane. The baseband building blocks of the frequency-locked loop are all implemented beneath the antenna using this area, and an additional unused area of 1mm² is available beneath the antenna for other circuits.

Acknowledgements:
The IOP sensor was designed by graduate students Gregory Chen, David Fick, Hassan Ghaed, Razi-ul Haque, Daeyeon Kim, Gyouho Kim, Yejoong Kim, Mingoo Seok, and post-doctoral researcher Michael Wieckowski.  The antenna-referenced frequency-locked loop was designed by graduate student Kuo-Ken Huang.