Abstract: An optical receiver includes an error generator, a multipath interference estimator, and a combiner. The error generator is configured to receive an input comprising a received optical signal, to estimate a modulation level of samples of the received optical signal, and to generate an error signal based on the estimated modulation level of the samples, the error signal representing a difference between an actual level of the received optical signal and the estimated modulation level. The multipath interference estimator is configured to generate estimates of multipath interference (MPI) associated with the samples of the received optical signal based on the error signal. The combiner is configured to generate an MPI-mitigated signal based on a combination of the samples and the estimates of MPI.

Abstract: A method and device for data conversion in a matrix compute apparatus. The apparatus includes an input buffer (IB) that receives one or more matrix inputs characterized by a first format and a compute device coupled to the IB device that is configured to determine a combined matrix output. The compute device determines a first matrix output using a first input portion of the matrix input and determines a second matrix output using a second input portion. The compute device then determines a combined matrix output in a second format using the first and second matrix outputs. Within the compute device, an alignment device can determine a rounded matrix output from the combined matrix output, and a partial products reduction (PPR) device can determine a reduced matrix output in a third format using the rounded matrix output, which is stored in an output buffer (OB) coupled to the compute device.

Abstract: An immersion cooling server system with AI accelerator apparatuses using in-memory compute chiplet devices. This system includes one or more immersion tanks with heat transfer fluid and configured with at least a condenser device. A plurality of AI accelerator servers is immersed in the heat transfer fluid in a bottom portion of the tanks and is configured to process transformer workloads while cooled by the immersion cooling configuration. Each of the servers includes a plurality of multiprocessors each having at least a first server central processing unit (CPU) and a second server CPU, both of which are coupled to a plurality of switch devices. Each switch device is coupled to a plurality of AI accelerator apparatuses. The apparatus includes one or more chiplets, each of which includes a plurality of digital in-memory compute (DIMC) devices configured to perform high throughput matrix computations for transformer based models.

Abstract: A method for data transmission includes receiving a data stream from a host device, the data stream as received from the host device including encoded data, separating the encoded data in the data stream into first data blocks and second data blocks, and generating a first forward error correction (FEC) block. The first FEC block includes a first parity section and a first data section, the first parity section includes a first parity bit corresponding to the first data blocks and a second parity bit corresponding to the second data blocks, and the first data section includes the first data blocks and the second data blocks. The method further includes transmitting the first FEC block.

Abstract: A server system with AI accelerator apparatuses using in-memory compute chiplet devices. The system includes a plurality of multiprocessors each having at least a first server central processing unit (CPU) and a second server CPU, both of which are coupled to a plurality of switch devices. Each switch device is coupled to a plurality of AI accelerator apparatuses. The apparatus includes one or more chiplets, each of which includes a plurality of tiles. Each tile includes a plurality of slices, a CPU, and a hardware dispatch device. Each slice can include a digital in-memory compute (DIMC) device configured to perform high throughput computations. In particular, the DIMC device can be configured to accelerate the computations of attention functions for transformer-based models (a.k.a. transformers) applied to machine learning applications. A single input multiple data (SIMD) device configured to further process the DIMC output and compute softmax functions for the attention functions.

Abstract: An optical transmitter includes a first encoder, a first interleaver, a second encoder, a mapper, a second interleaver, and a frame generator. The first encoder is configured to encode data using a staircase code to generate first codewords. The first interleaver is configured to interleave the first codewords using convolutional interleaving to spread a transmission order of the first codewords. The second encoder is configured to encode the interleaved first codewords using a second code to generate second codewords. The mapper is configured to map the second codewords to transmit symbols. The second interleaver is configured to interleave the transmit symbols to distribute the transmit symbols between pilot symbols. The frame generator is configured to generate a transmit frame including the interleaved transmit symbols and the pilot symbols.

Abstract: An AI accelerator apparatus using in-memory compute chiplet devices. The apparatus includes one or more chiplets, each of which includes a plurality of tiles. Each tile includes a plurality of slices, a central processing unit (CPU), and a hardware dispatch device. Each slice can include a digital in-memory compute (DIMC) device configured to perform high throughput computations. In particular, the DIMC device can be configured to accelerate the computations of attention functions for transformer-based models (a.k.a. transformers) applied to machine learning applications. A single input multiple data (SIMD) device configured to further process the DIMC output and compute softmax functions for the attention functions. The chiplet can also include die-to-die (D2D) interconnects, a peripheral component interconnect express (PCIe) bus, a dynamic random access memory (DRAM) interface, and a global CPU interface to facilitate communication between the chiplets, memory and a server or host system.

Abstract: A circuit and method for mitigating multi-path interference in direct detection optical systems is provided. Samples of an optical signal having a pulse amplitude modulated (PAM) E-field are processed by generating a PAM level for each sample. For each sample, the sample is subtracted from the respective PAM level to generate a corresponding error sample. The error samples are lowpass filtered to produce estimates of multi-path interference (MPI). For each sample, one of the estimates of MPI is combined with the sample to produce an interference-mitigated sample.

Abstract: An AI accelerator apparatus using in-memory compute chiplet devices. The apparatus includes one or more chiplets, each of which includes a plurality of tiles. Each tile includes a plurality of slices, a central processing unit (CPU), and a hardware dispatch device. Each slice can include a digital in-memory compute (DIMC) device configured to perform high throughput computations. In particular, the DIMC device can be configured to accelerate the computations of attention functions for transformer-based models (a.k.a. transformers) applied to machine learning applications. A single input multiple data (SIMD) device configured to further process the DIMC output and compute softmax functions for the attention functions. The chiplet can also include die-to-die (D2D) interconnects, a peripheral component interconnect express (PCIe) bus, a dynamic random access memory (DRAM) interface, and a global CPU interface to facilitate communication between the chiplets, memory and a server or host system.

Abstract: A method for data transmission includes receiving a data stream from a host device, the data stream as received from the host device including encoded data, separating the encoded data in the data stream into first data blocks and second data blocks, and generating a first forward error correction (FEC) block. The first FEC block includes a first parity section and a first data section, the first parity section includes a first parity bit corresponding to the first data blocks and a second parity bit corresponding to the second data blocks, and the first data section includes the first data blocks and the second data blocks. The method further includes transmitting the first FEC block.

Abstract: An AI accelerator apparatus using in-memory compute chiplet devices. The apparatus includes one or more chiplets, each of which includes a plurality of tiles. Each tile includes a plurality of slices, a central processing unit (CPU), and a hardware dispatch device. Each slice can include a digital in-memory compute (DIMC) device configured to perform high throughput computations. In particular, the DIMC device can be configured to accelerate the computations of attention functions for transformer-based models (a.k.a. transformers) applied to machine learning applications, including generative AI. A single input multiple data (SIMD) device configured to further process the DIMC output and compute softmax functions for the attention functions.

Abstract: An AI accelerator apparatus using in-memory compute chiplet devices. The apparatus includes one or more chiplets, each of which includes a plurality of tiles. Each tile includes a plurality of slices, a central processing unit (CPU), and a hardware dispatch device. Each slice can include a digital in-memory compute (DIMC) device configured to perform high throughput computations. In particular, the DIMC device can be configured to accelerate the computations of attention functions for transformer-based models (a.k.a. transformers) applied to machine learning applications. A single input multiple data (SIMD) device configured to further process the DIMC output and compute softmax functions for the attention functions. The chiplet can also include die-to-die (D2D) interconnects, a peripheral component interconnect express (PCIe) bus, a dynamic random access memory (DRAM) interface, and a global CPU interface to facilitate communication between the chiplets, memory and a server or host system.

Abstract: An AI accelerator apparatus using in-memory compute chiplet devices. The apparatus includes one or more chiplets, each of which includes a plurality of tiles. Each tile includes a plurality of slices, a central processing unit (CPU), and a hardware dispatch device. Each slice can include a digital in-memory compute (DIMC) device configured to perform high throughput computations. In particular, the DIMC device can be configured to accelerate the computations of attention functions for transformer-based models (a.k.a. transformers) applied to machine learning applications. A single input multiple data (SIMD) device configured to further process the DIMC output and compute softmax functions for the attention functions. The chiplet can also include die-to-die (D2D) interconnects, a peripheral component interconnect express (PCIe) bus, a dynamic random access memory (DRAM) interface, and a global CPU interface to facilitate communication between the chiplets, memory and a server or host system.

Abstract: The present invention is directed to data communication. More specifically, an embodiment of the present invention provides an error correction system. Input data signals are processed by a feedforward equalization module and a decision feedback back equalization module. Decisions generated by the decision feedback equalization module are processed by an error detection module, which determines error events associated with the decisions. The error detection module implements a reduced state trellis path. There are other embodiments as well.

Abstract: The present invention is directed to data communication system and methods. More specifically, various embodiments of the present invention provide a communication interface that is configured to transfer data at high bandwidth using nDSQ format(s) over optical communication networks. In certain embodiments, the communication interface is used by various devices, such as spine switches and leaf switches, within a spine-leaf network architecture, which allows large amount of data to be shared among servers.

Abstract: A method for data transmission includes receiving a data stream from a host device, the data stream as received from the host device including encoded data, separating the encoded data in the data stream into first data blocks and second data blocks, and generating a first forward error correction (FEC) block. The first FEC block includes a first parity section and a first data section, the first parity section includes a first parity bit corresponding to the first data blocks and a second parity bit corresponding to the second data blocks, and the first data section includes the first data blocks and the second data blocks. The method further includes transmitting the first FEC block.

Abstract: A circuit and method for mitigating multi-path interference in direct detection optical systems is provided. Samples of an optical signal having a pulse amplitude modulated (PAM) E-field are processed by generating a PAM level for each sample. For each sample, the sample is subtracted from the respective PAM level to generate a corresponding error sample. The error samples are lowpass filtered to produce estimates of multi-path interference (MPI). For each sample, one of the estimates of MPI is combined with the sample to produce an interference-mitigated sample.

Abstract: An optical transmitter includes a first encoder, a first interleaver, a second encoder, a mapper, a second interleaver, and a frame generator. The first encoder is configured to encode data using a staircase code to generate first codewords. The first interleaver is configured to interleave the first codewords using convolutional interleaving to spread a transmission order of the first codewords. The second encoder is configured to encode the interleaved first codewords using a second code to generate second codewords. The mapper is configured to map the second codewords to transmit symbols. The second interleaver is configured to interleave the transmit symbols to distribute the transmit symbols between pilot symbols. The frame generator is configured to generate a transmit frame including the interleaved transmit symbols and the pilot symbols.

Abstract: A method for data transmission includes receiving a data stream from a host device, the data stream as received from the host device including encoded data, separating the encoded data in the data stream into first data blocks and second data blocks, and generating a first forward error correction (FEC) block. The first FEC block includes a first parity section and a first data section, the first parity section includes a first parity bit corresponding to the first data blocks and a second parity bit corresponding to the second data blocks, and the first data section includes the first data blocks and the second data blocks. The method further includes transmitting the first FEC block.

Abstract: Embodiments relate to the emulation of the effect of Forward Error Correction (FEC) codes, e.g., GF10 Reed Solomon (RS) FEC codes, on the bit error ratio (BER) of received Pseudo-Random Binary Sequences (PRBS) patterns. In particular, embodiments group errors into RS-FEC symbols and codewords in order to determine if the errors are correctable. By emulating the error correction capabilities of FEC codes in order to determine which errors are correctable by the code, embodiments afford a more accurate representation of the post-FEC BER of RS FEC codes from links carrying PRBS patterns. This FEC code emulation provides error correction statistics, for stand-alone use or for error correction in connection with Bit Error Rate Testers (BERTs).