GoKawiilThe transmission of information in our universe—from the astronomic to the atomic—is mainly photonic. Although most of our digital data travel through photonic waveguides, a far larger data stream is transmitted photonically in the free-space world. An efficient chip-to-world photonic interface—that is, the ability to convert between the time-bin modes of an integrated electro-optic processor and the spatial modes of free space—creates opportunities in communications and ranging3,5,6,7, additive manufacturing8, near-eye displays9,10, biomedical imaging11,12, machine learning13 and atom control for quantum information14. However, our current digital infrastructure struggles with the immense data streams from the real world, in which every resolvable pixel is a channel that must be processed15. A similar challenge exists in quantum computing, which requires photonic control and readout of millions of physical qubits to achieve fault tolerance16. Concurrently, photonic integrated circuits (PICs) have proliferated17 and demonstrated sophisticated functionalities, including light conditioning for atomic arrays14 and free-space displays, in-physics algorithms13 and deep co-learning at the edge18. As digital electronics become more intelligent, the chip-to-world optical interface becomes a crucial link in the digital intelligence value chain.
Despite this, the lack of a mode-efficient interface between the guided-wave modes of PICs and the continuous modes of free space has prevented their seamless and scalable use. Integrated waveguide systems possess a large number of time-bin modes due to rapid electro-optic and all-optical interactions19, but have a limited number of waveguide modes, with broadband, diffraction-limited input or output available only at the chip edge20. By contrast, free space offers a nearly unbounded number of spatial modes21 with slower temporal variations for many applications3,4,9,22. Although the total mode counts are similar, existing solutions fail to bridge this mismatch due to poor mode quality1, limited fields of view (FOV), slow scan rates or a lack of direct, scalable PIC integration3,9,11,23,24. The ideal solution requires the ability to project and scan a diffraction-limited, single-mode beam to: (1) a large number of resolvable beam-spots N in the far field, (2) with a high refresh rate, (3) from a limited footprint and (4) directly on the surface of a programmable photonic chip. Current beam-scanning architectures face a fundamental trade-off: tiled aperture devices3 and optical phased arrays5 offer programmability but suffer from diffraction-degraded beam quality, whereas continuous aperture scanners11,25 are constrained by inertial limits and integration challenges. Consequently, existing solutions lack the ability to project scannable, broadband, non-diffractive emission directly from the chip surface2 (refer to Supplementary Section 8 for detailed architectural comparisons).
Although it is challenging to distil a single figure of merit (FOM) that captures all facets of a laser scanning system’s size, weight, power and cost, the per-footprint-area resolvable spot count is a foundational metric that not only determines the number of devices per wafer, but also has downstream effects on all of these key performance metrics. For our analysis, we quantify performance by taking the product of this footprint-adjusted spot count and the refresh rate, yielding a net FOM in units of spots s–1 mm–2. This simplified metric provides a baseline for comparison among different technologies26. Conventional pupil-plane scanners require large apertures for high resolution, which leads to slow, high-power actuation that limits FOMs to between approximately 500,000 and 1 million spots s–1 mm–2. By contrast, focal-plane scanners decouple optical and mechanical dimensions, but the use of bulk components has limited their FOM to fewer than 50,000 spots s–1 mm–2. Thus, the focal-plane scanning approach has been hindered by the lack of a scalable, actuatable single-mode waveguide that can be integrated directly on a PIC.
Here we introduce a new class of integrated photonic devices—the photonic ski-jump—that overcomes these challenges and enables a scalable chip-to-world photonic interface (Fig. 1a). This device, fabricated on a 200-mm wafer in a volume complementary-metal–oxide–semiconductor (CMOS) foundry, is composed of a nanoscale optical waveguide embedded monolithically on a piezoelectrically actuated microcantilever with submicrogram mass, thickness of about 2 µm and a large out-of-plane curvature. The small mass and physical dimensions overcome the inertial limits of scanning fibres and break the FOM trade-offs of pupil plane scanners. The large upward curvature is achieved by engineering the directionality of the intrinsic material stress differential between the thin film layers of the cantilever bimorph27 (Supplementary Section 1)—an approach inspired by mechanical metamaterials28 and which has been demonstrated on other quantum photonic platforms29. This provides vertical, scannable, broadband optical emission from anywhere on a 200-mm wafer with mechanical resonances from about 1 kHz to over 100 kHz, which significantly enhance the scan speed and FOV. The submicrometre integrated waveguides simultaneously minimize the mass and emitted spot-size, resulting in a greater-than-1,000-fold FOM improvement over existing fibre scanners11,25 and a greater-than-50-fold FOM improvement over mature micro-electro-mechanical systems (MEMS) mirrors3,4 and acousto-optic deflectors23,26,30 (Fig. 1b).
Fig. 1: Ski-jump integrated photonics and comparisons to current state-of-the-art beam scanning technologies. a, (i) Existing piezoelectric POMPIC components14,31,32,33,34 enable fast photonic control and information processing over many time-bin modes on a scalable photonics platform. (VL) π is the voltage–length product to achieve a π phase shift. (ii) Photonic ski-jumps enable beam-scanning and time-to-space mode conversion directly from the surface of a photonic chip. (iii) Targets in the free-space world have many spatial modes with slow temporal evolution. b, Comparison of the photonic ski-jump with leading laser beam scanners as a function of footprint-adjusted pixel density and refresh rate. Footprint refers to the active beam-scanner device area. Data points (green circles) are obtained from a single ski-jump measured in vacuum at 1, 2, 5 and 10 volts peak-to-peak (V pp ; left to right). Lower-left inset: projection plane FOV is given by the scan angle θ for pupil plane scanners or scan distance d for focal plane scanners scaled by magnification M. Acousto-optic deflector data points are from refs. 23,26,30. The MEMS mirror data points are the highest-performing devices from table 4 in ref. 3. The scanning fibre is from ref. 25 and the thermal MEMS is from ref. 2. c, A diced, unreleased POMPIC wafer. Scale bar, ~5 cm. d, A 64 ski-jump array on a POMPIC. Scale bar, 1 mm. e, Photonic ski-jumps integrated with other POMPIC components. Scale bar, 1 mm. Full size image
Photonic ski-jumps are members of a unified family of active components on a CMOS-compatible piezo-opto-mechanical photonic integrated circuit (POMPIC) platform. Past works on this platform are shown in Fig. 1a and include tunable directional couplers31, phase shifters32,33, programmable Mach–Zehnder interferometers32,34 and tunable ring resonators14,32. This extensive process development kit allows for complex photonic processing upstream of the ski-jump on the same monolithic photonic platform (Fig. 1c–e). Ski-jumps are also cryogenically compatible for direct integration with solid-state qubit systems such as colour centres in diamond, which have been heterogeneously integrated onto the POMPIC platform for microwave and strain control35. This opens up new routes for the addressing and readout of spin qubits. Future integration with electro-optic thin films36 could enable 100 GHz modulation for the projection of subnanosecond optical pulses. The capabilities of the POMPIC platform combined with the chip-to-world projection capability of the ski-jump enables scalable photonic and quantum control on- and off-chip.