William M. Mellette

RotorNet: A Scalable, Low-complexity, Optical Datacenter Network

The ever-increasing bandwidth requirements of modern datacenters have led researchers to propose networks based upon optical circuit switches, but these proposals face significant deployment challenges. In particular, previous proposals dynamically configure circuit switches in response to changes in workload, requiring network-wide demand estimation, centralized circuit assignment, and tight time synchronization between various network elements--- resulting in a complex and unwieldy control plane. Moreover, limitations in the technologies underlying the individual circuit switches restrict both the rate at which they can be reconfigured and the scale of the network that can be constructed. We propose RotorNet, a circuit-based network design that addresses these two challenges. While RotorNet dynamically reconfigures its constituent circuit switches, it decouples switch configuration from traffic patterns, obviating the need for demand collection and admitting a fully decentralized control plane. At the physical layer, RotorNet relaxes the requirements on the underlying circuit switches---in particular by not requiring individual switches to implement a full crossbar---enabling them to scale to 1000s of ports. We show that RotorNet outperforms comparably priced Fat Tree topologies under a variety of workload conditions, including traces taken from two commercial datacenters. We also demonstrate a small-scale RotorNet operating in practice on an eight-node testbed.

Scaling Limits of MEMS Beam-Steering Switches for Data Center Networks

Transparent optical circuit switching can improve the aggregate bandwidth, scalability, and cost of data center networks provided, it can meet the performance requirements on switching speed, port count, and optical efficiency. Here, we examine the theoretical scaling limits of transparent nonblocking optical switches based on MEMS electrostatic tilt mirror devices. Using physical optics and electromechanics, we present a first principles analysis of how the response speeds of a set of canonical devices scale as a function of switch port count, crosstalk, and insertion loss. Our model indicates that the optimal actuator design (parallel plate versus vertically offset comb) and actuation method (digital versus analog) changes as a function of switch port count. It also suggests that conventional switch topologies do not allow a favorable tradeoff between switching speed and optical efficiency or crosstalk. However, high switching speeds can be achieved by multistage switch architectures such as the two examples we describe, a multiport wavelength switch and a wavelength-independent space switch.

Quantitative analysis and temperature-induced variations of moiré pattern in fiber-coupled imaging sensors.

Imaging fiber bundles can map the curved image surface formed by some high-performance lenses onto flat focal plane detectors. The relative alignment between the focal plane array pixels and the quasi-periodic fiber-bundle cores can impose an undesirable space variant moiré pattern, but this effect may be greatly reduced by flat-field calibration, provided that the local responsivity is known. Here we demonstrate a stable metric for spatial analysis of the moiré pattern strength, and use it to quantify the effect of relative sensor and fiber-bundle pitch, and that of the Bayer color filter. We measure the thermal dependence of the moiré pattern, and the achievable improvement by flat-field calibration at different operating temperatures. We show that a flat-field calibration image at a desired operating temperature can be generated using linear interpolation between white images at several fixed temperatures, comparing the final image quality with an experimentally acquired image at the same temperature.

Digital image processing for wide-angle highly spatially variant imagers

High resolution, wide field-of-view and large depth-of-focus imaging systems are greatly desired and have received much attention from researchers who seek to extend the capabilities of cameras. Monocentric lenses are superior in performance over other wide field-of-view lenses with the drawback that they form a hemispheric image plane which is incompatible with current sensor technology. Fiber optic bundles can be used to relay the image the lens produces to the sensors planar surface. This requires image processing to correct for artifacts inherent to fiber bundle image transfer. Using a prototype fiber coupled monocentric lens imager we capture single exposure focal swept images from which we seek to produce extended depth-of-focus images. Point spread functions (PSF) were measured in lab and found to be both angle and depth dependent. This spatial variance enforces the requirement that the inverse problem be treated as such. This synthesis of information allowed us to establish a framework upon which to mitigate fiber bundle artifacts and extend the depth-of-focus of the imaging system.

Planar waveguide LED illuminator with controlled directionality and divergence

We present a versatile illumination system where white light emitting diodes are coupled through a planar waveguide to periodically patterned extraction features at the focal plane of a two dimensional lenslet array. Adjusting the position of the lenslet array allows control over both the directionality and divergence of the emitted beam. We describe an analytic design process, and show optimal designs can achieve high luminous emittance (1.3x10⁴ lux) over a 2x2 foot aperture with over 75% optical efficiency while simultaneously allowing beam steering over ± 60° and divergence control from ± 5° to fully hemispherical output. Finally, we present experimental results of a prototype system which validate the design model.

61 Port 1×6 selector switch for data center networks

We present design and preliminary characterization of a scalable MEMS-based “selector switch” for high performance computing networks. The 170 μs, 61-port prototype uses relay image steering to route all 61 SMF channels through one of six pre-structured interconnects.

P-FatTree: A multi-channel datacenter network topology

The bandwidth and latency requirements of next-generation datacenter networks stress the limits of CMOS manufacturing. A key trend in their design will be a move from single-channel links and switches to multi-channel links and switches. Todays network topologies erase this distinction, providing the illusion of a unified network fabric. In this work we propose P-FatTree, which is a FatTree topology designed specifically for the future multi-channel reality. P-FatTree requires fewer switch chips and as a result has lower cost, power consumption, and latency than existing approaches. Furthermore, by embracing the parallel nature of the network itself, it enables compelling new ways to better manage and deliver application traffic.

Panoramic full-frame imaging with monocentric lenses and curved fiber bundles

Panoramic imaging is important for many different applications, including content for immersive virtual reality. Although compact 360 cameras can be made from an array of small-aperture ‘smartphone’ imagers, their small (typically 1.1 m) pixels provide low dynamic range. Moreover, digital single-lens-reflex and cinematographic imagers have 4–8 m pixels, but require correspondingly longer focal length lenses. Conventional ‘fisheye’ lenses are also problematic because they are bulky and have low light collection (typically F/2.8 to F/4, where F is the focal length divided by the lens aperture). An alternative path to panoramic imaging is ‘monocentric’ optics, where all surfaces—including the image surface—are concentric hemispheres.1 The symmetry of these lenses means that lateral color and off-axis aberrations (astigmatism and coma) are eliminated. In addition, the simple lens structures can be used to correct for spherical and axial color aberrations to yield extraordinarily wide angle resolution and light collection.2 The image that is produced can be coupled to a conventional focal plane, via a fiber bundle faceplate (with a curved input and flat output face).3 Fiber faceplates are solid glass elements made of small, high-index optical fibers separated by a thin, low-index cladding, used for nonimaging transfer of light between the input and output faces. From our research, within the Defense Advanced Research Projects Agency (DARPA) SCENICC (Soldier Centric Imaging via Computational Cameras) program, we have shown that fiber bundles can reach a spatial resolution of 2 m.4 We have also Figure 1. Geometry of a monocentric lens (left) and the spherical image surface it forms (right) can be coupled to CMOS focal plane(s) by an array of straight fiber bundles (top) or a single curved fiber bundle (bottom). The F-number is the focal length (f) divided by the lens aperture.

Scaling limits of free-space tilt mirror MEMS switches for data center networks

We present a first-principles analysis of the scaling of switch response speed as a function of port count, crosstalk, and insertion loss, based on physical optics and kinematics of canonical MEMS tilt mirror switch structures.

Planar Waveguide Illuminator with Variable Directionality and Divergence

We present the design, model, and experimental characterization of a white light LED illuminator using mechanical actuation of a lenslet array relative to a micro-structured planar waveguide to control the divergence and direction of emitted light.