Related Experiment Video
Updated: Jun 13, 2026
07:56
A Photonic System for Generating Unconditional Polarization-Entangled Photons Based on Multiple Quantum Interference
Published on: September 5, 2019
An all-optical signal processor enabling terabit-per-second real-time equalization
Benshan Wang1, Qiarong Xiao1, Tengji Xu1
1Department of Electronic Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong SAR, China.
Science (New York, N.Y.)
|June 11, 2026
Summary
This study introduces an optical signal processor (OSP) for faster, more efficient AI training interconnects. The OSP offers real-time, all-optical equalization, significantly reducing latency and power consumption in graphics processing unit clusters.
Area of Science:
- Optical Engineering
- Artificial Intelligence Hardware
- Telecommunications
Background:
- Large-scale AI training requires high-speed, low-latency, energy-efficient interconnects for GPU clusters.
- Digital signal processing (DSP) equalization in intensity-modulation/direct-detection links faces limitations due to phase loss, power, and latency constraints.
Purpose of the Study:
- To present an integrated, programmable optical signal processor (OSP) for DSP-free, real-time equalization.
- To overcome the limitations of current DSP equalization methods in high-performance computing interconnects.
Main Methods:
- Development of an OSP featuring a deep reservoir with all-optical readout for a Vernier scheme.
- Implementation of ~1-picosecond sampling resolution and a tunable memory window.
- Simultaneous equalization of eight wavelength-division-multiplexing (WDM) channels.
Main Results:
- Achieved 1.6-terabits/second aggregate throughput with <60-picoseconds latency.
- Demonstrated energy consumption in the tens of femtojoules/bit.
- Provided superior chromatic dispersion compensation and mitigation of transceiver bandwidth limits and fiber nonlinearity.
Conclusions:
- The OSP enables all-optical, DSP-free, real-time equalization, functioning as a nonlinear universal equalizer.
- Expanded the usable WDM window by a factor of 6.8, enhancing optical link performance.
- Offers a promising solution for ultralow-latency, energy-efficient interconnects crucial for AI training.
Related Concept Videos
01:26Inductively Coupled Plasma Atomic Emission Spectroscopy: Instrumentation
Inductively coupled plasma (ICP) is the common plasma source used in atomic emission spectroscopy (AES), a technique that detects and analyzes various elements in a sample. This method is often called inductively coupled plasma atomic emission spectroscopy (ICP-AES).
There are three main types of inductively coupled plasma atomic emission spectroscopy (ICP-AES) instruments: sequential, simultaneous multichannel, and Fourier transform instruments, with the latter being less commonly used.
There are three main types of inductively coupled plasma atomic emission spectroscopy (ICP-AES) instruments: sequential, simultaneous multichannel, and Fourier transform instruments, with the latter being less commonly used.
01:10Reconstruction of Signal using Interpolation
Signal processing techniques are essential for accurately converting continuous signals to digital formats and vice versa. When a continuous signal is sampled with a period T, the resulting sampled signal exhibits replicas of the original spectrum in the frequency domain, spaced at intervals equal to the sampling frequency. To handle this sampled signal, a zero-order hold method can be applied, which creates a piecewise constant signal by retaining each sample's value until the next sampling...
01:18Aliasing
Accurate signal sampling and reconstruction are crucial in various signal-processing applications. A time-domain signal's spectrum can be revealed using its Fourier transform. When this signal is sampled at a specific frequency, it results in multiple scaled replicas of the original spectrum in the frequency domain. The spacing of these replicas is determined by the sampling frequency.
If the sampling frequency is below the Nyquist rate, these replicas overlap, preventing the original signal...
If the sampling frequency is below the Nyquist rate, these replicas overlap, preventing the original signal...
01:22Upsampling
Managing signal sampling rates is essential in digital signal processing to maintain signal integrity. A decimated signal, characterized by a reduced frequency range due to its lower sampling rate, can be upsampled by inserting zeros between each sample. This upsampling process expands the original spectrum and introduces repeated spectral replicas at intervals dictated by the new Nyquist frequency. To refine this zero-inserted sequence, it is passed through a lowpass filter with a cutoff...
01:29Downstream Processing
Downstream processing begins once fermentation is complete and involves a series of steps to recover and purify products such as acids, vitamins, antibiotics, or proteins.Cell HarvestingFor example, for intracellular protein-based products, the first step is harvesting the cells. This is typically achieved using centrifugation or filtration to separate the cells from the liquid phase.Cell Disruption for Intracellular ProductsIf the target product is intracellular, the harvested cells must be...