 |
Jerry McGuire Convergence Class Processors
What do I mean by a blended approach? I mean that a processing engine for applications with sophisticated control and communications (i.e. all convergence-class applications) must also be innately capable of executing powerful signal processing that is completely integrated within the application. And in the end, the only companies who are going to be able to successfully produce convergence-class processors are chip vendors who have a signal processing history, vendors who truly understand the needs of the application. "The reason grounding in signal processing is a non-negotiable need is that in convergence-class applications, control and signal processing are as fundamentally intertwined as a double helix — you can not partition convergence-class applications."The reason grounding in signal processing is a non-negotiable need is that in convergence-class applications, control and signal processing are as fundamentally intertwined as a double helix — you can not partition convergence-class applications. And if you can't partition the applications, you certainly can't partition your hardware into hybrid parts consisting of MCUs and DSP cores. And if you try, the results are far from optimal. You will be saddled with the costs of two development teams, the difficulty of maintenance of two development tool chains, the product costs of two cores and their respective memory systems and the inflexibility of an architecture with limited balance of control and signal processing capabilities. The central maxim of convergence Darwinism is that any embedded processor not intrinsically architected from the ground up for homogeneous, fully-integrated, high-performance signal processing and high-performance application processing, is one that embedded engineers will label as unfit for their designs. Processor architectures for convergence class applications must mirror the application demands because today's equipment users expect integrated features like intelligent sensing and reporting, high-speed communications, screamingly fast computing functions, outstanding graphics, secure and seamless wireless broadband, strong cryptographic security, and gorgeous streaming media (e.g., H.264 decode at 30 fps). Thus, convergence-class processors have to be able to handle seriously high-speed data streams over broadband communications channels with support for high-end processing applications and high-speed network data streams and manage increasingly natural human interfaces, inter-device communications, and system control. And don't eat my batteries — I want full D1 on my cell phone to cost me just 30 mW, please-and-thank-you. Look, users want easier interaction with every hidden computer they encounter, and they encounter more hidden computers every day. They want a rich experience, and the burden of this increased complexity demand falls directly on the convergence-class processor. Algorithms, bandwidth, and I/O. Plenty of chip companies have tried to "add media" to standard processors, but without heritage in signal processing it's just not going to happen. Here at ADI convergence class equals Blackfin Class, processors with enormous capacity for signal processing and control processing, processors suited for your convergent designs.
Jerry McGuire is a Vice President of Analog Devices' General Purpose DSP Group. His column appears here regularly. |
The confluence of embedded systems and signal-processing-intensive features is unmistakably defining the visible horizon and beyond. This so-called convergence class of applications is highly evolved from the primordial swamps of early embedded processing with bolt-on DSPs. Convergence-class applications, with their high-performance network-connectivity and highly-sophisticated human-machine interfaces, demand a truly blended approach to embedded computing and signal processing.