Artificial intelligence (AI) hardware is positioned to unlock revolutionary computational abilities by leveraging vast distributed networks of advanced semiconductor chips. However, a barrier for AI scaling is the disproportionately high energy and chip area required to transmit data between the chips. Here we present a solution to this long-standing overhead through dense three-dimensional (3D) integration of photonics and electronics. With 80 photonic transmitters and receivers occupying a combined chip footprint of only 0.3 mm², our platform achieves an order-of-magnitude-greater number of 3D-integrated channels than prior demonstrations. This enables both high-bandwidth (800 Gb s⁻¹) and highly efficient, dense (5.3 Tb s⁻¹ mm⁻²) 3D channels. The transceiver energy efficiency is showcased by a state-of-the-art 50 fJ and 70 fJ per communicated bit from the transmitter and receiver front ends, respectively, operating at 10 Gb s⁻¹per channel. Furthermore, the design is compatible with commercial complementary metal–oxide–semiconductor foundries fabrication on 300-mm-sized wafers, providing a route to mass production. Such ultra-energy-efficient, high-bandwidth data communication links promise to eliminate the bandwidth bottleneck between spatially distinct compute nodes and support the scaling of future AI computing hardware.

Significance: The smallest objects resolvable by the unaided human eye are approximately 100 μm in size. This limit sets a boundary between the familiar and the microscopic, the visible and invisible worlds. Here we describe the fabrication of tiny wireless sensors, optical wireless integrated circuits (OWICs) for optical wireless integrated circuits. OWICs are truly microscopic in size, visible to the naked eye at best as a tiny, undifferentiated speck, yet they can sense their environment and report the information back to the macroscopic world. They have potential applications in many areas of science and technology, ranging from neuroscience to chemical sensing.

Recording neural activity in live animals in vivo with minimal tissue damage is one of the major barriers to understanding the nervous system. This paper presents the technology for a tetherless opto-electronic neural interface based on 180 nm CMOS circuits, heterogeneously integrated with an AlGaAs diode that functions as both a photo voltaic and light emitting diode. These microscale opto-electrically transduced electrodes (MOTEs) are powered by, and communicate through an optical interface, simultaneously enabling high temporal-resolution electrical measurements without a tether or a bulky RF coil. The MOTE presented here is 250 μm × 57 μm, consumes 1 μW of electrical power, and is capable of capturing and encoding neural signals before transmitting the encoded signals. The measured noise floor is as low as 15 μV $_RMS$ at a 15 KHz bandwidth.

Autonomous robots—systems where mechanical actuators are guided through a series of states by information processing units to perform a predesigned function—are expected to revolutionize everything from health care to transportation. Microscopic robots are poised for a similar revolution in fields from medicine to environmental remediation. A key hurdle to developing these microscopic robots is the integration of information systems, particularly electronics fabricated at commercial foundries, with microactuators. Here, we develop such an integration process and build microscopic robots controlled by onboard complementary metal oxide semiconductor electronics. The resulting autonomous, untethered robots are 100 to 250 micrometers in size, are powered by light, and walk at speeds greater than 10 micrometers per second. In addition, we demonstrate a microscopic robot that can respond to an optical command. This work paves the way for ubiquitous autonomous microscopic robots that perform complex functions, respond to their environments, and communicate with the outside world.

In vivo, chronic neural recording is critical to understand the nervous system, while a tetherless, miniaturized recording unit can render such recording minimally invasive. We present a tetherless, injectable micro-scale opto-electronically transduced electrode (MOTE) that is ~60μm x 30μm x 330μm, the smallest neural recording unit to date. The MOTE consists of an AlGaAs micro-scale light emitting diode (μLED) heterogeneously integrated on top of conventional 180nm complementary metal-oxide-semiconductor (CMOS) circuit. The MOTE combines the merits of optics (AlGaAs μLED for power and data uplink), and of electronics (CMOS for signal amplification and encoding). The optical powering and communication enable the extreme scaling while the electrical circuits provide a high temporal resolution (<100μs). This paper elaborates on the heterogeneous integration in MOTEs, a topic that has been touted without much demonstration on feasibility or scalability. Based on photolithography, we demonstrate how to build heterogenous systems that are scalable as well as biologically stable - the MOTEs can function in saline water for more than six months, and in a mouse brain for two months (and counting). We also present handling/insertion techniques for users (i.e. biologists) to deploy MOTEs with little or no extra training. [2020-0080]