Chips&Media CFrame60 is a next-generation frame compression hardware IP designed to address the growing bandwidth and memory challenges in modern SoCs targeting imaging, video, AI, and display applications. Unlike conventional compression architectures that prioritize either bandwidth reduction or image quality, CFrame60 introduces a flexible framework supporting both lossless and lossy compression while minimizing SRAM requirements and DRAM traffic. The architecture is optimized for ISP pipelines, NPUs, VPUs, GPUs, and display subsystems where power efficiency and latency are critical design constraints.
One of the key architectural advantages of CFrame60 is its ability to process both raster-scan and block-based pixel ordering. Traditional compression schemes such as AFBC and DSC often require large line buffers because the compressor and decompressor operate in different pixel orders than adjacent IP blocks. This mismatch forces the system designer to implement SRAM-heavy line conversion stages. In 4K YUV422 10-bit systems, these line buffers can exceed 160KB, significantly increasing silicon area and power consumption. CFrame60 eliminates this requirement by supporting native interoperability with raster-scan pipelines used in ISP and display engines as well as block-based ordering commonly found in video codecs.
The CFrame60 product family consists of three major configurations: CFrame60C, CFrame60V, and CFrame60R. CFrame60C targets general-purpose lossless and lossy compression for ISP, NPU, and display applications. CFrame60V extends the architecture to support video processing systems requiring mixed raster and block-mode operations. CFrame60R further adds random access and partial update capabilities for advanced memory-efficient workflows. This modular product segmentation enables SoC architects to optimize implementation cost according to workload requirements while maintaining software compatibility across the platform.
Random access functionality in CFrame60R is particularly important for modern imaging systems and AI-enhanced video pipelines. Conventional frame compression solutions typically require sequential decompression beginning from the start of the bitstream. In contrast, CFrame60R supports decompression at arbitrary Access Units (AUs), allowing direct retrieval of localized image regions. This capability dramatically reduces unnecessary memory reads and lowers latency in warping, distortion correction, and region-of-interest processing applications. Partial update support additionally allows modified AUs to be rewritten independently without recompressing the entire frame, making the architecture highly efficient for interactive graphics and dynamic overlays.
Another technical differentiator is the rate-control mechanism used in lossy mode. Many competing frame compression solutions rely on fixed bitrate allocation that can introduce visible artifacts in flat textures or high-frequency edge regions. CFrame60 incorporates image complexity detection to dynamically allocate compression resources based on local image characteristics. Internal benchmark comparisons against DSC 1.2, AFBC, and JPEG-XS demonstrate superior PSNR and SSIM metrics across standard DSC test sequences. In challenging textures such as fine text rendering, noise gradients, and sweep patterns, CFrame60 achieves visually lossless quality while maintaining compression ratios between 1/2 and 1/4.
The hardware implementation is also optimized for low-latency operation. CFrame60 requires only one to two lines of latency compared with competing architectures that may require 16 to 32 lines of buffering. This reduction directly improves responsiveness in real-time video systems and lowers memory subsystem pressure. The IP supports YUV400, YUV420, YUV422, YUV444, RGB, and Bayer formats with bit depths ranging from 8-bit to 16-bit, enabling deployment across camera, automotive, AI, and professional imaging markets. Standard interfaces include AXI4 and on-the-fly ready/valid streaming protocols, simplifying integration into heterogeneous SoC environments.
From a performance perspective, CFrame60 scales efficiently through multi-core operation. A single core supports 4K 60fps processing at frequencies ranging from 250MHz to 450MHz depending on chroma format, while eight-core implementations can scale up to 8K 120fps or 16K 30fps applications. Silicon area is also competitive, with encoder logic sizes ranging from approximately 85K to 165K gates depending on feature configuration and supported bit depth. Memory requirements remain comparatively small, contributing to reduced total SoC power consumption.
The broader significance of CFrame60 lies in its system-level efficiency. As AI-enhanced imaging, computational photography, and high-resolution video continue driving memory bandwidth growth, compression IP becomes increasingly critical to overall SoC scalability. By combining flexible processing order support, low SRAM dependency, high image quality, and random access capability, CFrame60 positions itself as a practical alternative to legacy display-oriented compression technologies. The roadmap toward CFrame70 further indicates expansion into ultra-high-ratio visually lossless compression targeting JPEG-XS-class applications while maintaining smaller silicon footprint and simplified operation models.
Also Read:
WAVE-N Specialized Video Processing NPU for Edge AI Systems
CEO Interview with Steve Kim of Chips&Media
Share this post via:
Comments
There are no comments yet.
You must register or log in to view/post comments.