top of page
oscarfallas

Leveraging Low Latency to the Next Level with the Holoscan Sensor Bridge


NVIDIA Holoscan Sensor Bridge
Source: Lattice Semiconductor. Lattice and NVIDIA Edge AI Solution.

The Holoscan Sensor Bridge is a board from Lattice Semiconductor and NVIDIA. The board provides low-latency sensor data processing using an FPGA interface and the Holoscan NVIDIA software. The main idea of the Holoscan is to provide the data directly into the host device's memory.


Hardware Composition


The Sensor Bridge board comprises three main parts, as described below.


  • Image Sensors and RF Sensors: Two camera ports capture data and use MIPI D-PHY 2.5G and the I2C protocol. It is also compatible with ultrasound, MR, CT, PET, and radio sensors.

  • Field Programmable Gate Arrays (FPGA): Lattice Semiconductor developed a cross-work between two FPGAs. The CrossLink is a low-power FPGA solution capable of handling MIPI D-PHY with support of 4K UHD resolution at speeds up to 12 Gbps. The CertusPro-NX is an advanced general-purpose FPGA that assists its users with rapid prototyping and testing. The main modules of these two FPGAs are I/O, Out of Band Module, Packetizer, and EN/ET.

  • SFP+ Transceivers: The two Ethernet ports work with enhanced form factor with data rates of 10Gbps.



NVIDIA Holoscan Block Diagram
Source: Lattice Semiconductor. Lattice and NVIDIA Edge AI Solution

Host Devices Supported


The Holoscan Sensor Bridge communicates the sensor with the host system at low latency. Currently, the board supports two different NVIDIA systems: NVIDIA IGX Orin and NVIDIA Jetson AGX Orin Developer Kit.


Software Composition


The Holoscan Core is an AI sensor processing platform for hardware systems for low-latency sensing and network connectivity.

  • The core is an application composed of fragments that are a building block of an app and can be assigned to a physical node of the Holoscan cluster.

  • The fragments are composed of operators, the most basic unit of the framework. The Operators are in charge of receiving and processing the streaming data with a main function.

  • The Operators are composed of inputs and outputs, letting the operators combine and convert different inputs into a specific output.


Holoscan Software API brings into two flavors for the user: Holoscan C++ API and Holoscan Python API


Use Cases for Low Latency


Nowadays, low latency devices are one of the most demanding fields of study due to the vast number of applications that need to work as quickly as possible to achieve the goal of why they were built. The main objective of the Holoscan Sensor Bridge is to work as a low-overhead image transmitter; some use cases are described below. 


  • Medical Devices/Industry: More specific and complex devices are needed as medical sciences advance. The low latency image capturing could improve areas such as faster medical examination results, real-time diagnosis, remote procedures with robotic assistants, or automatic medical procedures.


  • Exploration and Discovery: Space and underwater exploration are the areas with the more complex conditions. With faster image capturing and data analysis, first-time environments could reduce the risk of damage and increase success rates. 


  • Autonomous Vehicles: Lately, one of the most famous achievements has been the improvement of human quality of life by autonomous vehicles. The low latency environmental analysis is the most complex and important goal those devices could meet. 


  • Powerful Microservices and APIs: In a microservices environment, where applications are composed of small, independent services that communicate over a network, quick data analysis and decision-making are crucial. Reduced latency ensures faster response times for inter-service communication, leading to more efficient processing and reduced bottlenecks. 


Comparative with other solutions 


High-speed interfaces and low-latency applications have been approached with many solutions over the years. A device like a PCIe card, a peripheral connected to a computer's motherboard, creates a high-speed interface to graphics cards, network devices, and storage controllers. The Holoscan Sensor Bridge have some advantages over the PCle, such as: 


  • Physical Space: PCIe cards require a motherboard with compatible PCIe slots, which increases the physical space needed for device development and the hardware costs. In contrast, the Holoscan Sensor Bridge only needs an Ethernet connection between the host device and the card, making it much easier to accommodate and distribute in any space.


  • Power Consumption: High-performance PCIe cards, such as GPUs or high-speed network cards, consume significant amounts of power, necessitating robust power supplies and efficient cooling solutions. In contrast, FPGAs are known for their low power consumption, making the Holoscan Sensor Bridge a strong competitor in energy efficiency.


Other solutions involve the Jetson Bridges, part of NVIDIA's robotic application platform. They are similar to Holoscan Sensor Bridge, which works with an NVIDIA host device and uses AI software to optimize data. An advantage of the Holoscan Sensor Bridge over the Jetson Bridge is the use of Field Programmable Gate Arrays, which opens the possibility to improve or adapt the card to the specific goal for which the device is built. Besides, the Jetson Bridges are usually built for one specific task and are complex to adapt to another one.


Interest Research Areas 


Due to the adaptability of the Holoscan Sensor Bridge, there are a lot of research areas in which to use the device. The board is designed for low-latency data processing, which is ideal for developing real-time or Low-Latency applications. Another interest is using the board’s FPGAs as CPU accelerators by employing parallel processing, offloaded execution, and other advanced optimization methods. Artificial Intelligence is one of the most in-demand fields in the industry today. Utilizing FPGAs to enhance hardware performance and improve AI capabilities is another key area of study for the board.



Interested in Reaching the Next Level?



Special thanks to Oscar Fallas, author of this post.

Recent Posts

See All
bottom of page