HiPEAC’s Tech Transfer Awards recognize successful examples of technology transfer, which covers technology licensing, providing dedicated services or creating a new company, for example. All applications are evaluated by an internal technology transfer committee. In addition to a certificate, first-time winners are awarded the sum of €1,000 for the team that developed the technology.
This year, eight winners have been selected. In alphabetical order, they are as follows:
- Monica de Mier, Bytelab Solutions: Bytelab Solutions, a new company for the commercialization of atomistic simulations
- Daniel Hofman, University of Zagreb: Jobbee system
- Leonidas Kosmidis, Barcelona Supercomputing Center (BSC): Brook SC: High-level Certification-friendly Programming for GPU-powered Safety Critical Systems
- Filippo Mantovani, Barcelona Supercomputing Center (BSC): Data visualization for industry 4.0
- Farhad Merchant, RWTH Aachen University: Parameterized posit arithmetic hardware generator and its integration to a RISC-V based platform
- Tuan Nguyen, student of Akash Kumar, Technische Universität Dresden: Automatic Floorplanner for Partially Reconfigurable FPGA-based Design
- Horacio Pérez-Sánchez, José Pedro Cerón Carrasco, Jorge de la Peña García and Helena den Haan Alonso, Universidad Católica San Antonio de Murcia (UCAM): Computational drug discovery of clinically approved drugs as potent in vitro and in vivo Zika virus inhibitors
- Hans Salomonsson, EmbeDL: EmbeDL - Optimization of Deep Learning in Embedded Systems
Further information
Monica de Mier, Bytelab Solutions: Bytelab Solutions, a new company for the commercialization of atomistic simulations
Density functional theory (DFT) is a popular tool for the computational design and characterization of materials, thanks to the good compromise it offers between precision and speed. However, the high computational complexity of DFT means that systems with more than a few hundred atoms cannot be simulated.
The open-source code BigDFT, co-developed at Barcelona Supercomputing Center (BSC), overcomes this issue by implementing linear scaling algorithms that allow calculations for much bigger systems to be performed. The company Bytelab Solutions, a start-up dedicated to the acceleration of materials research and development via digitalization, is developing and commercializing a software platform that integrates this technology. The company’s mission is to democratize the usage of computer simulations for the faster (and thus more cost-efficient) development of materials.
Daniel Hofman, University of Zagreb: Jobbee system
Jobbee is an innovative, predictive, geo-location sensitive system for matching recruiters and candidates, tasks and skills, based on scientifically developed algorithms and inputs from human resources and human behaviour expertise. It was created thanks to the transfer of knowledge from the field of data management, data mining, recommendation systems, user tracking and parallel processing from the University of Zagreb to the private company VIDI-to.
While complex in the background, drawing on target groups, customer relationship management systems, human behaviour on the web (including social networks), and usage of smartphone/ tablet applications, Jobbee is designed to be easy for users. The tools are scaled to explore Jobbee user habits and make decisions based on them, while the delivery of information to the users uses recommendation systems and personalization methodologies able to scale to the needs of the ICT system.
Leonidas Kosmidis, Barcelona Supercomputing Center (BSC): Brook SC: High-level Certification-friendly Programming for GPU-powered Safety Critical Systems
Critical systems, such as those used in the automotive and avionics sectors, require increased performance in order to support advanced functionalities such as autonomous operation. Graphics processing units (GPUs) can provide the required level of performance; however, their general programming models, such as CUDA or OpenCL, cannot be used in such systems as they violate safety critical programming guidelines.
Brook SC was developed to allow safety-critical applications to be programmed in a CUDA-like high-level general-purpose GPU language, Brook, which enables the certification of the code while increasing developers’ productivity.
In this technology transfer, an Airbus prototype application performing general-purpose computations with a safety-critical graphics API was ported to use Brook SC in record time. This achieved an order of magnitude reduction in the lines of code to implement the same functionality without a penalty in performance.
Filippo Mantovani, Barcelona Supercomputing Center (BSC): Data visualization for industry 4.0
Aingura IIoT provides solutions to improve productivity through smart factory operational architectures. These include solutions for the engineering company Etxetar – which produces machinery for the automotive sector – for collecting, synchronizing, fusing and analysing industrial data, with the aim of reducing machine downtime.
In this collaboration, the BSC visualization tool Paraver, which analyses performance, was transferred to Aingura IIoT. Normally used for high-performance computing (HPC) applications, in accordance with Aingura IIoT’s needs Paraver can also be used to:
- represent time-stamped data series such as values from sensors
- combine data series so that, for example, power drain from current intensity and voltage can be compared
- visualize the data in graphs, so that, for example, outliers in the data – such as malfunctioning sensors – can be detected
Farhad Merchant, RWTH Aachen University: Parameterized posit arithmetic hardware generator and its integration to a RISC-V based platform
Posit arithmetic has been proposed as an alternative to the traditional floating point standard IEEE 754 for computing. It offers advantages such as higher dynamic range, better accuracy and numerical precision, a lower hardware footprint, and lower power requirements.
RWTH Aachen worked with the Bosch Research and Technology Center in Bangalore, India on two technology transfers:
- Developing a parameterized posit arithmetic hardware generator, named PAU generator
- Integration of PAU-generated hardware onto a RISC-V platform
It is projected that Bosch will be able to source 10-12% more microprocessors for the cameras when enabled with posit arithmetic This technology is also expected to support applications such as intelligent video analytics (IVA), motion+ and future generation electronic control units (ECUs) in the next few years.
Tuan Nguyen, student of Akash Kumar, Technische Universität Dresden: Automatic Floorplanner for Partially Reconfigurable FPGA-based Design
One of the main benefits of field-programmable gate arrays (FPGAs) is the ability to reconfigure small areas of the FPGA without interrupting the whole system. As a result, the system can be dynamically updated and loaded with new features at runtime.
However, to enable partial reconfiguration, FPGA designers have to floorplan the design on the FPGA. This requires not only thorough knowledge of the FPGA but also the design architecture to be able to find suitable locations on the FPGA for the partially reconfigurable regions. Currently, automatic floorplanning for partial reconfiguration is not offered by commercial electronic design and automation (EDA) vendors.
The floorplanning technology developed in this project, which was transferred from TU Dresden to Huawei, is able to automatically floorplan the design within minutes, even with a very large design of which resources requirements are up to 85% of the FPGA.
Horacio Pérez-Sánchez, José Pedro Cerón Carrasco, Jorge de la Peña García and Helena den Haan Alonso, Universidad Católica San Antonio de Murcia (UCAM): Computational drug discovery of clinically approved drugs as potent in vitro and in vivo Zika virus inhibitors
Members of the BIO-HPC group at UCAM created a marketable solution using advanced computational drug discovery technologies to predict compounds that could inhibit the Zika virus. Following tests at the University of Hong Kong, one of the predicted compounds, novobiocin, was found to potentially inhibit the Zika virus. A patent was obtained and subsequently licensed to the pharmaceutical company Ennaid Therapeutics. Clinical trials will begin in 2020.
Hans Salomonsson, EmbeDL: EmbeDL - Optimization of Deep Learning in Embedded Systems
EmbeDL is a research spinout that can help organizations bring deep learning (DL)-based artificial intelligence (AI) into physical products (embedded systems). EmbeDL can make AI components in cars, drones and even smart homes significantly faster while using less energy and memory.
EmbeDL was developed by the AI research and Innovation company Machine Intelligence Sweden AB in the Horizon 2020 financed project LEGaTO, which aims to build a new software stack for low energy computing on heterogeneous hardware. At the heart of this technology is the DL optimization engine developed in LEGaTO. The EmbeDL deep learning optimization engine uses components from the project to support a great variety of compute platforms.
EmbeDL takes a DL model developed in one of the popular frameworks, e.g. Tensorflow, and transforms the model to a new model significantly more efficient on a target hardware. The process is fully automated.