Deep Learning Processor
Flexible architecture for deep learning and computer vision
The videantis v-MP6000UDX architecture runs a wide variety of visual processing tasks very efficiently, providing a flexible visual computing platform that runs complete embedded vision applications that combine CNNs with computer vision, video compression, and image processing. The heterogeneous multicore architecture includes multiple high-throughput VLIW/SIMD media processors (v-MPs) with a number of stream processors (v-SPs) that accelerate bitstream packing and unpacking in video codecs. Each processor includes its own multi-channel DMA engine for efficient data movement to local, on-chip, and off-chip memories.
Low power extends battery life and prevents overheating
Low energy consumption extends battery life of mobile devices and also prevents camera modules and image sensors from overheating in automotive and edge computing applications. A new class of computer vision applications, such as face recognition for smart sensing devices, require always-on operation, again necessitating ultra-low power consumption. The v-MP6000UDX processor has been architected for low-power operation from the ground up, resulting in a 1000x power reduction compared to other solutions on the market.
High-performance enables new applications
Deep learning, computer vision applications, 4K/8K or Full HD video coding, and high-resolution image processing require tremendous amounts of computing power. With performance of 60TOPS (Tera Operations Per Second; example for 256 v-MP cores implemented in a 7nm process), the v-MP6000UDX processor architecture provides ample computational horsepower. Of course the videantis processing platform can be scaled up even further to fulfil the highest performance requirements. The heterogeneous, multicore VLIW/SIMD architecture was designed specifically for visual computing tasks such as neural network processing, computer vision algorithms and complex video applications, resulting in at least a 100x performance improvement compared to other processors.
Single scalable architecture
The first videantis multicore architecture was introduced at the Microprocessor Forum 2005 and since then underwent continuous improvements. Also, for more than a decade videantis has worked on optimizing and parallelizing software for efficient multicore operation. The single scalable architecture enables different operating points across product lines: from very-low-cost to ultra-high-performance. The videantis software is fully scalable too, automatically using all processor cores that are available for a given task. The v-CNNDesigner tool efficiently parallelizes neural networks over multiple cores, ensuring there are no data bottlenecks and guaranteeing high data throughput. Available videantis embedded vision and video coding libraries also take full advantage of the all the cores available in the system.
Software programmable
The videantis v-MP6000UDX processor architecture is fully software programmable and thus can handle the still rapidly changing deep learning algorithms. Neural networks are fully implemented in software – optimization can still continue after deployment because a changed network structure or new functionality can be accommodated by a firmware upgrade. Therefore a SoC based on the videantis v-MP6000UDX processing platform can stay in the market longer or can even address other market segments with a wide variety of customer requests.
Production proven
The videantis processors have been designed into a wide variety of SoCs and millions of products have shipped into the market. Available videantis-based SoCs as well as videantis tools and development processes fulfil the highest requirements for functional safety according to ISO26262. Several generations of videantis' processing platform have been licensed since 2004, helping to build a superb track record of processor technology and support.
View Deep Learning Processor full description to...
- see the entire Deep Learning Processor datasheet
- get in contact with Deep Learning Processor Supplier
Block Diagram of the Deep Learning Processor IP Core
Processor IP
- RISC-V ARC-V RMX-100 Ultra-low Power 32-bit Processor IP for Embedded Applications
- ARC-V RMX-500 power efficient 32-bit RISC-V processor for embedded applications
- ARC-V RHX-105 dual-issue, 32-bit RISC-V processor for real-time applications (multi-core)
- Secure-IC Securyzr(TM) Cyber Escort Unit IP provides real time detection of sero day attacks on processor
- 64-bit RISC-V Application Processor Core
- 64-bit RISC-V Application Processor Core