Rapid prototyping platform for AI processors
This software solution can be used for quick architecture exploration of electronics and, embedded software that implements Artificial Intelligence (AI) and Machine Learning. This AI evaluator accelerates the deployment of new generation devices by providing a large library of components to quickly prototype, simulate, analyze and optimize the processing, communication and storage. Example of application prototype constructed with this environment include 5G base-stations, mobiles, autonomous driving, medical systems, Intelligent processors, Industrial IoT and consumer electronics.
A unique aspect of this prototyping platform is that it incorporates all aspects of the system including the software, accelerators, cache structures, processor, memories, buses, networks, SSD, MEMS, RF, analog and peripherals. VisualSim AI Designer can help identify areas for AI acceleration, define the AI architecture including the memory access, bus interface and the application partitioning across multiple processing tiles, and compare different AI approaches and semiconductors. This prototype can become the specification for product development and the platform for early software definition and development.
Unlike an FPGA prototype or a software model, the VisualSim prototype can be probed to view the system internals such as buffer usage, resource efficiency, system reliability, power consumption and security features. Also, these VisualSim prototypes are built prior to any development, thus gaining insight into project schedule, resource requirements, potential verification zones and early demonstrator. VisualSim models are constructed in a graphical environment using pre-defined parameterized building blocks, simulated, and analyzed using over 500 statistical reports.
Using VisualSim AI Evaluator, architects and though-leaders can design custom accelerators, partition the application into AI, processor and FPGA, and determine the implementation, topology, system sizing and memory access. The design criteria are based on use-cases and traffic input; and the generated statistics provides insight into deadlocks, system bottlenecks and requirements matching.
This VisualSim AI Evaluator contains four different modelling technologies, a Discrete-Event Simulation with 90 million events per second, interactive visualization and human-readable documentation output. The modelling technologies include:
1. Library of analog components, memories, queues, buses, protocols, multi-core, GPUs, FPGAs, ASICs, CPUs, accelerators, and High-Performance Computing (HPC) structures
2. Task graph modelling components to emulate the application
3. Analysis tools for task tracing, deadline monitors, memory access, data overload and efficiency
4. Use-case and traffic generators to emulate the environment and input streams.
VisualSim AI Designer is fully integrated with the industry-standard VisualSim Architect.
Mirabilis Design – www.mirabilisdesign.com
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
