News

ROS 2 Hardware Acceleration Working Group #14

Source: Acceleration Robotics, January 15, 2023 [1]
Join the exciting ROS 2 Hardware Acceleration Working Group on its 14th meeting to discuss how hardware acceleration and custom compute architectures can speed up robotics.

The goal of this meeting is to review the progress we have made over the last year, report on the current status of the group, and discuss our goals for the upcoming year.

Acceleration robotics joins AESEMI to lead new processor architectures

Source: Acceleration Robotics, January 16, 2023 [2]
The robotics semiconductor company joins a dozen other organizations to position Spanish talent on the international semiconductor scene. It will lead the development of new computing architectures and seeks collaborations in robotics.

ROS 2 hardware acceleration working group (HAWG) in 2022

Source: Acceleration Robotics, January 15, 2023 [3]
The ROS 2 Hardware Acceleration Working Group (HAWG) launched new projects in 2022 including RobotPerf, the Robotics MCU or the Robotics Processing Unit and achieved a 3.7x growth in outreach over 2021, engaging with more than 1M roboticists within the robotics WG activities during the last year.

How human language accelerated robotic learning

Source: The Robot Report, January 12, 2023 [4]
Princeton researchers have found that human-language descriptions of tools can accelerate the learning of a simulated robotic arm lifting and using a variety of tools. The results build on evidence that providing richer information during artificial intelligence training can make autonomous robots more adaptive to new situations, improving their safety and effectiveness.

How Nvidia’s CUDA monopoly in machine learning is breaking

Source: SemiAnalysis, January 16, 2023 [5]
Nvidia's dominant position in machine learning is being disrupted. PyTorch 2.0 and OpenAI's Triton are replacing CUDA as the default for training AI models. This report examines why Google's TensorFlow failed and how Nvidia's advantage in CUDA is being wiped away.

2022 started with a bang but ended with a whimper for semiconductor companies

Source: The Register, January 17, 2023 [6]
Worldwide semiconductor revenues grew just 1.1 percent during 2022, a far cry from a year ago when the annual increase was more than 25 percent. Memory was the worst-performing market segment, declining 10 percent against the previous year, and worse may be to come in 2023. Worldwide semiconductor revenues grew just 1.1 percent during 2022, a far cry from a year ago.

Taiwan's chip exports rose as China's imports fell in 2022

Source: The Register, January 16, 2023 [7]
Taiwan continues to dominate the global semiconductor manufacturing industry. Exports of semiconductors from Taiwan went up 18.4 percent year-on-year in 2022. China's imports of integrated circuits fell for the first time in many years, as US sanctions bite.

TSMC reports fourth quarter financial results

Source: HPCwire, January 12, 2023 [8]
Year-over-year, fourth quarter revenue increased 42.8% while net income and diluted EPS both increased 78.0%.

How motion engineering helps develop next-gen surgical robots

Source: The Robot Report, January 18, 2023 [9]
Next-generation motion engineering can help you develop the next generation of surgical robots. Surgical robots will enable surgeons to perform less invasive, more precise operations. The ideal angle of approach for the camera and instruments into the incision site is as parallel and close together as possible.

Why are embedded systems so far behind?

Source: embedded, January 18, 2023 [10]
Embedded systems still need to catch up with other areas of technology in terms of code development. Microcontrollers (MCUs) have restricted resources, which might restrict the development and application possibilities. As the IoT develops, there will be a greater demand for embedded devices with cutting-edge connection and capabilities.

Selecting the right RISC-V core

Source: Semiconductor Engineering, January 16, 2023 [11]
The RISC-V design and software ecosystems for processors are becoming established. End users face an increasingly difficult challenge of ensuring they make the best choices. The flexibility of Risc-V creates huge challenges for verification, above any beyond what is required for verification of fixed processors.

Power integrity analysis for high-performance FPGAs

Source: Semiconductor Engineering, January 12, 2023 [12]
Efinix field-programmable gate arrays (FPGAs) are custom-tailored for the computing demands of mainstream applications. Customers use the Titanium line of FPGAs to ensure their complex, high-performance designs minimize power consumption. They must seamlessly integrate IP such as microcontrollers, serial peripheral interfaces and inter-integrated circuit controllers.

Source: NVIDIA Technical Blog, January 17, 2023 [13]
CUDA Toolkit 12.0 introduces a new nvJitLink library for Just-in-Time Link Time Optimization (JIT LTO) support. Previously, developers had to build and compile CUDA kernels as a single source file in whole programming mode. In some cases, the performance gain was reported to be 20% or more.

Papers

Duet: Creating harmony between processors and embedded FPGAs

Source: arXiv, January 7, 2023 [14]
Duet is a scalable, manycore-FPGA architecture that promotes embedded FPGAs to be equal peers with processors through non-intrusive, bi-directionally cache-coherent integration. Duet can reduce the processor-accelerator communication latency by up to 82% and increase the bandwidth by 9.5x.

PiDRAM: A holistic end-to-end FPGA-based framework for processing-in-DRAM

Source: ACM Transactions on Architecture and Code Optimization, November 17, 2022 [15]
Processing-in-DRAM is the first flexible end-to-end framework that enables system integration studies and evaluation of real, commodity DRAM-based PuM techniques. We implement PiDRAM on an FPGA-based RISC-V system. Integrating D-RaNGe and RowClone takes only 388 lines of Verilog and 643 lines of C++ code.


Previous Hardware Acceleration in Robotics Newsletters

Past ROS 2 Hardware Acceleration Working Group meetings


  1. https://www.linkedin.com/events/ros2hardwareaccelerationworking7020416834277392384/about/ ↩︎

  2. Mayoral-Vilches, V. (2023, January 17). Acceleration robotics joins AESEMI to lead new processor architectures. Acceleration Robotics. https://news.accelerationrobotics.com/acceleration-robotics-joins-aesemi-lead-processor-architectures/ ↩︎

  3. ROS 2 hardware acceleration working group (HAWG) in 2022. (2023, January 16). Hardware Acceleration in Robotics. https://news.accelerationrobotics.com/ros-2-hardware-acceleration-working-group-2022-dissemination-report/ ↩︎

  4. The Robot Report Staff. (2023, January 12). How human language accelerated robotic learning. The Robot Report. https://www.therobotreport.com/how-human-language-accelerated-robotic-learning/?spMailingID=82262&puid=2502618&E=2502618&utm_source=newsletter&utm_medium=email&utm_campaign=82262 ↩︎

  5. Patel, D. (2023, January 16). How Nvidia’s CUDA monopoly in machine learning is breaking - OpenAI Triton and PyTorch 2.0. SemiAnalysis | Dylan Patel | Substack. https://www.semianalysis.com/p/nvidiaopenaitritonpytorch ↩︎

  6. Robinson, D. (2023, January 17). Global semiconductor revenues grew just 1.1% in 2022. The Register: Enterprise Technology News and Analysis. https://www.theregister.com/2023/01/17/global_semiconductor_revenues_2022/?utm_source=daily&utm_medium=newsletter&utm_content=article ↩︎

  7. Robinson, D. (2023, January 16). Taiwan's chip exports rose as China's imports fell in 2022. The Register: Enterprise Technology News and Analysis. https://www.theregister.com/2023/01/16/taiwan_chip_exports_rose_last/?utm_source=daily&utm_medium=newsletter&utm_content=article ↩︎

  8. TSMC reports fourth quarter financial results. (2023, January 12). HPCwire. https://www.hpcwire.com/off-the-wire/tsmc-reports-fourth-quarter-financial-results/ ↩︎

  9. Crowe, S. (2023, January 18). How motion engineering helps develop next-Gen surgical robots. The Robot Report. https://www.therobotreport.com/motion-engineering-helps-develop-next-gen-surgical-robots/ ↩︎

  10. Rabault, N. (2023, January 18). Why are embedded systems so far behind? embedded. https://www.embedded.com/why-are-embedded-systems-so-far-behind/ ↩︎

  11. Bailey, B. (2023, January 16). Selecting the right RISC-V core. Semiconductor Engineering. https://semiengineering.com/selecting-the-right-risc-v-core/?cmid=d17e87d7-68c1-4348-877e-f0bfc7564120 ↩︎

  12. Wagnon, J. (2023, January 12). Power integrity analysis for high-performance FPGAs. Semiconductor Engineering. https://semiengineering.com/power-integrity-analysis-for-high-performance-fpgas/?cmid=0523777e-2817-49ce-a7bf-7c23b9ef28b3 ↩︎

  13. Sundaram, A., Murphy, M., Ferrer, M., Ligowski, L., & Oh, F. (2023, January 17). CUDA 12.0 compiler support for runtime LTO using nvJitLink library. NVIDIA Technical Blog. https://developer.nvidia.com/blog/cuda-12-0-compiler-support-for-runtime-lto-using-nvjitlink-library/ ↩︎

  14. Ang, L., Ning, A., & Wentzlaff, D. (2023, January 7). Duet: Creating harmony between processors and embedded FPGAs. arXiv.org. https://arxiv.org/abs/2301.02785 ↩︎

  15. Olgun, A., Gomez, J., Kanellopoulos, K., Salami, B., Hassan, H., Ergin, O., & Mutlu, O. (2022, November 17). PiDRAM: A holistic end-to-end FPGA-based framework for processing-in-DRAM. ACM Transactions on Architecture and Code Optimization. https://dl.acm.org/doi/full/10.1145/3563697 ↩︎