Roadmap on emerging hardware and technology for machine learning.

Recent progress in artificial intelligence is largely attributed to the rapid development of machine learning, especially in the algorithm and neural network models. However, it is the performance of the hardware, in particular the energy efficiency of a computing system that sets the fundamental limit of the capability of machine learning. Data-centric computing requires a revolution in hardware systems, since traditional digital computers based on transistors and the von Neumann architecture were not purposely designed for neuromorphic computing. A hardware platform based on emerging devices and new architecture is the hope for future computing with dramatically improved throughput and energy efficiency. Building such a system, nevertheless, faces a number of challenges, ranging from materials selection, device optimization, circuit fabrication and system integration, to name a few. The aim of this Roadmap is to present a snapshot of emerging hardware technologies that are potentially beneficial for machine learning, providing the Nanotechnology readers with a perspective of challenges and opportunities in this burgeoning field.

Single-chip microprocessor that communicates directly using light.

Data transport across short electrical wires is limited by both bandwidth and power density, which creates a performance bottleneck for semiconductor microchips in modern computer systems--from mobile phones to large-scale data centres. These limitations can be overcome by using optical communications based on chip-scale electronic-photonic systems enabled by silicon-based nanophotonic devices. However, combining electronics and photonics on the same chip has proved challenging, owing to microchip manufacturing conflicts between electronics and photonics. Consequently, current electronic-photonic chips are limited to niche manufacturing processes and include only a few optical devices alongside simple circuits. Here we report an electronic-photonic system on a single chip integrating over 70 million transistors and 850 photonic components that work together to provide logic, memory, and interconnect functions. This system is a realization of a microprocessor that uses on-chip photonic devices to directly communicate with other chips using light. To integrate electronics and photonics at the scale of a microprocessor chip, we adopt a 'zero-change' approach to the integration of photonics. Instead of developing a custom process to enable the fabrication of photonics, which would complicate or eliminate the possibility of integration with state-of-the-art transistors at large scale and at high yield, we design optical devices using a standard microelectronics foundry process that is used for modern microprocessors. This demonstration could represent the beginning of an era of chip-scale electronic-photonic systems with the potential to transform computing system architectures, enabling more powerful computers, from network infrastructure to data centres and supercomputers.

Optimization of photoluminescence from W centers in a silicon-on-insulator.

W centers are trigonal defects generated by self-ion implantation in silicon that exhibit photoluminescence at 1.218 µm. We have shown previously that they can be used in waveguide-integrated all-silicon light-emitting diodes (LEDs). Here we optimize the implant energy, fluence and anneal conditions to maximize the photoluminescence intensity for W centers implanted in silicon-on-insulator, a substrate suitable for waveguide-integrated devices. After optimization, we observe near two orders of magnitude improvement in photoluminescence intensity relative to the conditions with the stopping range of the implanted ions at the center of the silicon device layer. The previously demonstrated waveguide-integrated LED used implant conditions with the stopping range at the center of this layer. We further show that such light sources can be manufactured at the 300-mm scale by demonstrating photoluminescence of similar intensity from 300 mm silicon-on-insulator wafers. The luminescence uniformity across the entire wafer is within the measurement error.

Deuterated silicon nitride photonic devices for broadband optical frequency comb generation.

We report and characterize low-temperature, plasma-deposited deuterated silicon nitride films for nonlinear integrated photonics. With a peak processing temperature less than 300°C, it is back-end compatible with complementary metal-oxide semiconductor substrates. We achieve microresonators with a quality factor of up to 1.6×106 at 1552 nm and >1.2×106 throughout λ=1510-1600 nm, without annealing or stress management (film thickness of 920 nm). We then demonstrate the immediate utility of this platform in nonlinear photonics by generating a 1 THz free-spectral-range, 900 nm bandwidth modulation-instability microresonator Kerr comb and octave-spanning, supercontinuum-broadened spectra.

Room-temperature-deposited dielectrics and superconductors for integrated photonics.

We present an approach to fabrication and packaging of integrated photonic devices that utilizes waveguide and detector layers deposited at near-ambient temperature. All lithography is performed with a 365 nm i-line stepper, facilitating low cost and high scalability. We have shown low-loss SiN waveguides, high-Q ring resonators, critically coupled ring resonators, 50/50 beam splitters, Mach-Zehnder interferometers (MZIs) and a process-agnostic fiber packaging scheme. We have further explored the utility of this process for applications in nonlinear optics and quantum photonics. We demonstrate spectral tailoring and octave-spanning supercontinuum generation as well as the integration of superconducting nanowire single photon detectors with MZIs and channel-dropping filters. The packaging approach is suitable for operation up to 160 °C as well as below 1 K. The process is well suited for augmentation of existing foundry capabilities or as a stand-alone process.

Ultra-low-loss CMOS-compatible waveguide crossing arrays based on multimode Bloch waves and imaginary coupling.

We experimentally demonstrate broadband waveguide crossing arrays showing ultralow loss of 0.0 4dB/crossing (0.9%) on average and converging to 0.033 dB/crossing (0.075%) matching theory and cross-talk suppression over 35 dB in a CMOS-compatible geometry. The principle of operation is the tailored excitation of a low-loss spatial Bloch wave formed by matching the periodicity of the crossing array to the difference in propagation constants of the first- and third-order TE-like modes of a multimode silicon waveguide. Radiative scattering at the crossing points acts like a periodic imaginary-permittivity perturbation that couples two supermodes, which results in imaginary (radiative) propagation-constant splitting and gives rise to a low-loss, unidirectional breathing Bloch wave. This type of crossing array provides a robust implementation of a key component enabling dense photonic integration.

Polycrystalline silicon ring resonator photodiodes in a bulk complementary metal-oxide-semiconductor process.

We present measurements on resonant photodetectors utilizing sub-bandgap absorption in polycrystalline silicon ring resonators, in which light is localized in the intrinsic region of a p+/p/i/n/n+ diode. The devices, operating both at λ=1280 and λ=1550 nm and fabricated in a complementary metal-oxide-semiconductor (CMOS) dynamic random-access memory emulation process, exhibit detection quantum efficiencies around 20% and few-gigahertz response bandwidths. We observe this performance at low reverse biases in the range of a few volts and in devices with dark currents below 50 pA at 10 V. These results demonstrate that such photodetector behavior, previously reported by Preston et al. [Opt. Lett. 36, 52 (2011)], is achievable in bulk CMOS processes, with significant improvements with respect to the previous work in quantum efficiency, dark current, linearity, bandwidth, and operating bias due to additional midlevel doping implants and different material deposition. The present work thus offers a robust realization of a fully CMOS-fabricated all-silicon photodetector functional across a wide wavelength range.

Depletion-mode carrier-plasma optical modulator in zero-change advanced CMOS.

We demonstrate the first (to the best of our knowledge) depletion-mode carrier-plasma optical modulator fabricated in a standard advanced complementary metal-oxide-semiconductor (CMOS) logic process (45 nm node SOI CMOS) with no process modifications. The zero-change CMOS photonics approach enables this device to be monolithically integrated into state-of-the-art microprocessors and advanced electronics. Because these processes support lateral p-n junctions but not efficient ridge waveguides, we accommodate these constraints with a new type of resonant modulator. It is based on a hybrid microring/disk cavity formed entirely in the sub-90 nm thick monocrystalline silicon transistor body layer. Electrical contact of both polarities is made along the inner radius of the multimode ring cavity via an array of silicon spokes. The spokes connect to p and n regions formed using transistor well implants, which form radially extending lateral junctions that provide index modulation. We show 5 Gbps data modulation at 1265 nm wavelength with 5.2 dB extinction ratio and an estimated 40 fJ/bit energy consumption. Broad thermal tuning is demonstrated across 3.2 THz (18 nm) with an efficiency of 291 GHz/mW. A single postprocessing step to remove the silicon handle wafer was necessary to support low-loss optical confinement in the device layer. This modulator is an important step toward monolithically integrated CMOS photonic interconnects.

Depletion-mode polysilicon optical modulators in a bulk complementary metal-oxide semiconductor process.

We demonstrate depletion-mode carrier-plasma optical modulators fabricated in a bulk complementary metal-oxide semiconductor (CMOS), DRAM-emulation process. To the best of our knowledge, these are the first depletion-mode modulators demonstrated in polycrystalline silicon and in bulk CMOS. The modulators are based on novel optical microcavities that utilize periodic spatial interference of two guided modes to create field nulls along waveguide sidewalls. At these nulls, electrical contacts can be placed while preserving a high optical Q. These cavities enable active devices in a process with no partial silicon etch and with lateral p-n junctions. We demonstrate two device variants at 5 Gbps data modulation rate near 1610 nm wavelength. One design shows 3.1 dB modulation depth with 1.5 dB insertion loss and an estimated 160 fJ/bit energy consumption, while a more compact device achieves 4.2 dB modulation depth with 4.0 dB insertion loss and 60 fJ/bit energy consumption. These modulators represent a significant breakthrough in enabling active photonics in bulk silicon CMOS--the platform of the majority of microelectronic logic and DRAM processes--and lay the groundwork for monolithically integrated CMOS-to-DRAM photonic links.

Considerations for Neuromorphic Supercomputing in Semiconducting and Superconducting Optoelectronic Hardware.

Any large-scale spiking neuromorphic system striving for complexity at the level of the human brain and beyond will need to be co-optimized for communication and computation. Such reasoning leads to the proposal for optoelectronic neuromorphic platforms that leverage the complementary properties of optics and electronics. Starting from the conjecture that future large-scale neuromorphic systems will utilize integrated photonics and fiber optics for communication in conjunction with analog electronics for computation, we consider two possible paths toward achieving this vision. The first is a semiconductor platform based on analog CMOS circuits and waveguide-integrated photodiodes. The second is a superconducting approach that utilizes Josephson junctions and waveguide-integrated superconducting single-photon detectors. We discuss available devices, assess scaling potential, and provide a list of key metrics and demonstrations for each platform. Both platforms hold potential, but their development will diverge in important respects. Semiconductor systems benefit from a robust fabrication ecosystem and can build on extensive progress made in purely electronic neuromorphic computing but will require III-V light source integration with electronics at an unprecedented scale, further advances in ultra-low capacitance photodiodes, and success from emerging memory technologies. Superconducting systems place near theoretically minimum burdens on light sources (a tremendous boon to one of the most speculative aspects of either platform) and provide new opportunities for integrated, high-endurance synaptic memory. However, superconducting optoelectronic systems will also contend with interfacing low-voltage electronic circuits to semiconductor light sources, the serial biasing of superconducting devices on an unprecedented scale, a less mature fabrication ecosystem, and cryogenic infrastructure.

Broad tuning of whispering-gallery modes in silicon microdisks.

Silicon microdisks with dynamically-tunable resonance spectra are achieved with nanoscale, in-plane silicon electrical contacts in a single lithographic step. Electrical current is passed through the devices to enable thermal tuning via joule heating. A 14nm wavelength shift is demonstrated with 1.6mW power consumption in devices with >20nm free spectral ranges and quality factors exceeding 20,000. Spectral shifts equal to a full width at half maximum can be achieved with approximately 10microW tuning power for a mode with quality factor of 20,000.

Continuous-wave subwavelength microdisk lasers at λ = 1.53 µm.

Subwavelength InGaAs/AlInAs microdisk lasers are demonstrated under continuous-wave optical pumping at a heat-sink temperature of 45 K. A 1.49 µm diameter, 209 nm thick microdisk lases in single-mode at a wavelength of 1.53 µm, which is identified as the whispering-gallery mode with the first radial mode number, the fifth azimuthal mode number, and a modal volume of 2.12(λ/n)(3) according to our mode simulation.

Slow light and band gaps in metallodielectric cylinder arrays.

We consider two-dimensional three-component photonic crystals wherein one component is modeled as a drude-dispersive metal. It is found that the dispersion relation of light in this environment depends critically on the configuration of the metallic and dielectric components. In particular, for the case of an incident electromagnetic wave with electric field vector parallel to the axis of the cylinders it is shown that the presence of dielectric shells covering the metallic cylinders leads to a closing of the structural band gap with increased filling factor, as would be expected for a purely dielectric photonic crystal. For the same polarization, the photonic band structure of an array of metallic shell cylinders with dielectric cores do not show the closing of the structural band gap with increased filling factor of the metallic component. In this geometry, the photonic band structure contains bands with very small values of group velocity with some bands having a maximum of group velocity as small as .05c.

Electroluminescence of nanopatterned silicon with carbon implantation and solid phase epitaxial regrowth.

Electroluminescence at 1.28mum is observed in a nanopatterned silicon test structure that has been subjected to carbon implantation followed by solid-phase epitaxial regrowth for recrystalization. The sub-bandgap luminescence comes from a di-carbon complex known as 'G center'. Enrichment of silicon with carbon atoms has been achieved in a procedure consisting of two implantations and solid-phase epitaxial regrowth. Nanopatterning was done using an anodized aluminum oxide membrane as a mask for reactive ion etching. Along with the electroluminescence, an enhanced photoluminescence was measured.

All-silicon light-emitting diodes waveguide-integrated with superconducting single-photon detectors.

We demonstrate cryogenic, electrically injected, waveguide-coupled Si light-emitting diodes (LEDs) operating at 1.22 µm. The active region of the LED consists of W centers implanted in the intrinsic region of a p-i-n diode. The LEDs are integrated on waveguides with superconducting nanowire single-photon detectors (SNSPDs). We demonstrate the scalability of this platform with an LED coupled to eleven SNSPDs in a single integrated photonic device.