Contact Sales

Overview

Our Hummingbird™ processor utilizes advanced vertically stacked packaging technologies to integrate a photonic and electronic die into one single package. Its revolutionary architecture provides a path forward to solve the “memory wall” faced by chip architects designing novel System-in-Packages with NoC interconnects. Serving as the communications network for data centers and other high-performance applications, Hummingbird is the first in a family of products that utilize Lightelligence’s oNOC platform, improving computing performances by enabling innovative interconnect topologies via silicon photonics. Its waveguides propagate signals at the speed of light and utilize an all-to-all data broadcast network to each core on a 64-core domain-specific AI processor chip, giving Hummingbird significant advantages in latency and power reduction over traditional digital interconnect solutions. Unlike digital networks, Hummingbird’s oNOC technology increases density scaling by enabling interconnect topologies that would be otherwise unrealizable. It enables higher utilization of computing power due to more efficient communication. With oNOC, mapping workloads to hardware becomes easier and provides greater freedom to select the right topology for the computing task. Hummingbird is implemented as a low-latency optical all-to-all broadcast network spanning 64 cores. With 64 transmitters and 512 receivers, it provides a framework to implement a variety of dense optical network topologies. Hummingbird's electronic and photonic ICs are co-packaged and integrated into a PCIe form factor for installation in industry-standard servers. Coupled with the Lightelligence Software Development Kit (SDK), machine learning and AI workloads can be optimized to take full advantage of the oNOC.
Click to expand

Key Technology

Computing power improvement have relied on the increase in density of transistors, referred to Moore’s law. However, Moore’s law is slowing down by reaching physical limits and performance improvement being slower than transistor density. A new compute paradigm beyond Moore’s law is needed and it’s no longer just about density, it’s about scaling silicon area. Hummingbird is the world’s 1st compute product revealing the next architecture necessary to scale computing.

Cross sectional view of electrical chips interconnected
through optical waveguides

Top view of wafer level optical network formed by interconnecting
electrical chips through optics

Software Stack

Adaptable with your existing
TensorFlow models

Provides a heterogenous computing platform

Provides a heterogenous computing platform

Natively supports nearly all TensorFlow operators

Natively supports nearly all TensorFlow operators

Accelerates targeted graphs and dedicated operators

Accelerates targeted graphs and dedicated operators

3D Packaging Architecture


Hummingbird uses advanced vertically stacked packaging technologies to integrate a photonic IC and an electric IC into one single package.

Electronic Chip

EIC: Digital domain includes key parts of INS/CTL, DPMP, ALU and UMEM. Analog domain coordinate with PIC to implement oNOC.

Interposer

Interposer: Connect EIC, PIC and substrate for data transfer and power supply.

Photonic Chip

PIC: Optical Network-on-Chip (oNOC) to implement broadcasting network among all computation units.

Technology File

The information has been submitted, thank you!

Ok