Search for: All records

Abstract High-bandwidth applications, from multi-gigabit communication and high-performance computing to radar signal processing, demand ever-increasing processing speeds. However, they face limitations in signal sampling and computation due to hardware and power constraints. In the microwave regime, where operating frequencies exceed the fastest clock rates, direct sampling becomes difficult, prompting interest in neuromorphic analog computing systems. We present the first demonstration of direct broadband frequency domain computing using an integrated circuit that replaces traditional analog and digital interfaces. This features a Microwave Neural Network (MNN) that operates on signals spanning tens of gigahertz, yet reprogrammed with slow, 150 MBit/sec control bitstreams. By leveraging significant nonlinearity in coupled microwave oscillators, features learned from a wide bandwidth are encoded in a comb-like spectrum spanning only a few gigahertz, enabling easy inference. We find that the MNN can search for bit sequences in arbitrary, ultra-broadband10 GBit/sec digital data, demonstrating suitability for high-speed wireline communication.Notably, it can emulate high-level digital functions without custom on-chip circuits, potentially replacing power-hungry sequential logic architectures. Its ability to track frequency changes over long capture times also allows for determining flight trajectories from radar returns. Furthermore, it serves as an accelerator for radio-frequency machine learning, capable of accurately classifying various encoding schemes used in wireless communication. The MNN achieves true, reconfigurable broadband computation, which has not yet been demonstrated by classical analog modalities, quantum reservoir computers using superconducting circuits, or photonic tensor cores, and avoidsthe inefficiencies of electro-optic transduction. Its sub-wavelength footprint in a Complementary Metal-Oxide-Semiconductor process and sub-200 milliwatt power consumption enable seamless integration as a general-purpose analog neural processor in microwave and digital signal processing chips. 

Increasingly functional microscopic machines are poised to have massive technical influence in areas including targeted drug delivery, precise surgical interventions, and environmental remediation. Such functionalities would increase markedly if collections of these microscopic machines were able to coordinate their function to achieve cooperative emergent behaviors. Implementing such coordination, however, requires a scalable strategy for synchronization—a key stumbling block for achieving collective behaviors of multiple autonomous microscopic units. Here, we show that pulse-coupled complementary metal-oxide semiconductor oscillators offer a tangible solution for such scalable synchronization. Specifically, we designed low-power oscillating modules with attached mechanical elements that exchange electronic pulses to advance their neighbor’s phase until the entire system is synchronized with the fastest oscillator or “leader.” We showed that this strategy is amenable to different oscillator connection topologies. The cooperative behaviors were robust to disturbances that scrambled the synchronization. In addition, when connections between oscillators were severed, the resulting subgroups synchronized on their own. This advance opens the door to functionalities in microscopic robot swarms that were once considered out of reach, ranging from autonomously induced fluidic transport to drive chemical reactions to cooperative building of physical structures at the microscale. 

We present a tunable LNA for software defined radio based on a compact, tunable transmission line (CTTL) element. The CTTL acts as a passive, widely tunable LC resonance in a cascoded, common source LNA to implement an instantaneously narrowband, multi-octave tunable LNA. The resulting circuit, fabricated in 65nm CMOS, is tunable from 3.5-20GHz, and consumes 12 mW with gain >12dB, ≥ -9.6dBV in-band OP1dB, and OOB B1dB up to 31dB higher than the in-band B1dB due to the CTTL-tuned LC filtering. The CTTL-tuned LNA represents a more blocker-tolerant approach to achieving high frequency, software-defined LNAs without significant compromises in other LNA performance metrics. 

Cognitive radio aims at identifying unused radio-frequency (RF) bands with the goal of re-using them opportunistically for other services. While compressive sensing (CS) has been used to identify strong signals (or interferers) in the RF spectrum from sub-Nyquist measurements, identifying unused frequencies from CS measurements appears to be uncharted territory. In this paper, we propose a novel method for identifying unused RF bands using an algorithm we call least matching pursuit (LMP). We present a sufficient condition for which LMP is guaranteed to identify unused frequency bands and develop an improved algorithm that is inspired by our theoretical result. We perform simulations for a CS-based RF whitespace detection task in order to demonstrate that LMP is able to outperform black-box approaches that build on deep neural networks.