Marvell Asia

A method and system for determining, providing, and utilizing enhanced telemetry data in a computer system or network. The enhanced telemetry data includes access latency for a region or subregion of memory and information about the memory type, memory access type, and memory distance of a operation utilizing that region or subregion of memory. The enhanced telemetry data is transmitted in an expanded acknowledgement packet used by commutation protocols such as the UET protocol. The enhanced telemetry data provides increased detail and valuable information regarding that takes into account the varying access and performance times of different operations and locations to improve system commutations, task management, and troubleshooting. In particular, the enhanced telemetry data allows for the selection of delivery contexts between two nodes to avoid head of line blocking.

A glass optical interposer assembly (11, 90) includes: (i) a glass pane (14) having one or more optical waveguides (16) embedded therein, (ii) a compute die (70) coupled to the glass pane (14), (iii) a plurality of high bandwidth memory (HBM) components (72) coupled to the glass pane (14) and arranged around the compute die (70), and (iv) a plurality of optical transceiver integrated circuits (22) embedded within the glass pane (14) and positioned at a depth having a Z dimension offset relative to respective ones of the compute die (70) and the plurality of HBM components (72), the one or more optical waveguides (16) provide optical communication pathways between the compute die (70) and the plurality of HBM components (72) through the plurality of optical transceiver integrated circuits (22).

Methods for communicating in a computing network and corresponding devices are described. An example method of communicating in a computing network includes, by a scale-up switch, receiving a command packet generated by a compute component of a computing network, the command packet including a destination identifier and a memory address. The method further includes identifying a destination compute component based on the command packet received and determining if the destination compute component is in a scale-up network or a scale-out network. Responsive to the destination compute component being in the scale-out network, the method includes translating the command packet received into a network message, the translating including mapping the destination identifier to a network address and the memory address to a network resource index, and transmitting the network message to a scale-out fabric. Said communicating in the computing network can reduce demands upon the compute component and streamline network communication.

Methods for communication in a computing network and accompanying scale-up switches and design structures are described herein. According to an example embodiment, a method of communication in a computing network includes, by a scale-up switch, receiving a command packet generated by a compute component of a computing network. The method further includes, by the scale-up switch, identifying a destination compute component based on the command packet received and determining if the destination compute component is in a scale-up network or a scale-out network of the computing network. Responsive to the destination compute component being in the scale-out network, the method further includes, by the scale-up switch, embedding the command packet received in a network message and transmitting the network message to a scale-out fabric of the scale-out network. Communication in the computing network as described can be helpful for reducing demands upon the compute component and streamlining network communication.

A laser device based on silicon photonics with an in-cavity power monitor includes a gain chip (100) mounted on a silicon photonics substrate and configured to emit light in an active region bounded between a frontend facet (101) with low reflectivity and a backend facet (102) with anti-reflective characteristics. The laser device further includes a wavelength tuner (200) formed with waveguides (210) in the silicon photonics substrate optically coupled to the backend facet to receive light from the gain chip and configured to have a reflector (201) with high reflectivity to reflect the light in an extended cavity formed with the frontend facet through which a laser with a tuned wavelength and amplified power is outputted. Additionally, the laser device includes a photodiode (205) formed in the silicon photonics substrate and coupled to the waveguides in the extended cavity right in front of the reflector to measure power of light thereof.

An electro-optical system includes an optical source, an optical modulator, a first photodetector (PD), a second PD, and a control circuit. The optical source generates an optical signal. The optical modulator modulates the optical signal. The first PD detects the optical signal at an input to the optical modulator. The second PD detects the modulated optical signal at an output of the optical modulator. The control circuit sends to the optical modulator a modulation bias signal that includes a tone, determines a DC power ratio between the modulated optical signal detected by the second PD, and the optical signal detected by the first PD, and sets a bias of the optical modulator according to both (i) the DC power ratio and (i) an amplitude of a second harmonic of the tone in the modulated optical signal at the output of the optical modulator.

An optical device for enabling more flexible and precise control of spectral reflectance in a selected wavelength range having a maximal wavelength (λ_max) and a minimal wavelength (λ_min). The device includes a first Mach-Zehnder interferometer (MZI), including first and second legs having a tunable first optical pathlength difference (OPD) greater than 2×λ_max, and a second MZI, coupled in series with the first MZI and including third and fourth legs having a tunable second OPD less than 2×λ_min. An optical coupler is configured to direct an input optical signal in the selected wavelength range into the first and second MZIs and to output a reflected optical signal from the first and second MZIs. A controller is configured to adjust the first and second tunable OPDs to control the spectral reflectance of the reflected output optical signal.

A pilot-assisted fiber length estimation (FLE) technique estimates chromatic dispersion for optical signals. The pilot-assisted FLE technique is agnostic to the modulation format of the optical payload and is resilient to polarization effects, bandwidth limitations, and signal shaping, thus enabling consistent performance across a wide range of operating conditions. A transmitter periodically inserts a known pilot sequence into an optical signal, and a receiver processes the pilot sequence to estimate the length of the optical fiber, or to estimate chromatic dispersion of the optical fiber, which is related to the fiber length.

Inline cryptography techniques implemented with a storage accelerator may enable encryption of data in transit between the storage accelerator and the storage device. In one example, the storage accelerator maintains local I/O context information for I/O commands and modifies the data pointer (e.g., PRP) in the I/O command sent to the storage device to include a reference (e.g., index) to the I/O context information. For a read command, the storage accelerator receives encrypted data from the SSD along with the modified data pointer. The storage accelerator may then determine the appropriate key and tweak value based on the modified data pointer, decrypt the data with the key and tweak value, and send the decrypted data to the host device without waiting for all data to be received from the storage device.

A PHY device for transmitting data over a communication channel comprises a distribution circuit and a failure management circuit. The distribution circuit is configured to receive data from a host via a first plurality of lanes at a first data rate, and distribute the data into a second plurality of lanes for transmission over the communication channel. The failure management circuit is configured to monitor the communication channel, and detect failure of one of the second plurality of lanes. Subsequent to the detected failure, the distribution circuit is configured to receive the data from the host through the first plurality of lanes at a second data rate that is less than the first data rate, and to distribute the data received from the host through the first plurality of lanes at the second data rate into remaining lanes of the second plurality of lanes at the first data rate.