Cadence’s Tensilica team was honored with the Best Paper Award at the IP track at the Design Automation Conference last week for “Design in the Eye of the Hurricane—Building Optimal Vision Processing Systems.” The paper highlighted the technology behind the Tensilica IVP (Image/Vision Processing) DSP. Cadence IP Group CTO Chris Rowen accepted the award (pictured) on behalf of the team which included Greg Efland, Shrinivas Gadkari, Vadim Kustov, Dan Nicolaescue, Himanshu Sanghavi, and Sandip Parikh. "The IVP family is highly efficient and versatile to handle the innate diversity and computational complexity of vision processing," Rowen said. "It’s proven to be suitable across a wide range of noise reduction, object recognition, convolutional neural network, and other complex vision, video, and imaging functions." This paper is very timely, as the demands for vision applications are exploding. Just look at advanced driver assistance systems (ADAS). You need vision DSPs in the back-up camera, auto-dimming headlights, blind spot detection, parking assist, rain/fog detection, pedestrian detection and avoidance, vehicle detection and collision avoidance, and many other applications. When looking for a vision instruction set, Rowen advised that people look for: High local memory bandwidth Effective latency hiding for DDR access 8b, 16b, and 32b fixed- and floating-point data types Vision-specific operations such as 2D data access, convolution search, histogram, and more Automatic compiler inference of vectors and complex operations Be sure to check out the Tensilica IPV DSP.![]()
![]()
↧