Analog Devices latest Blackfin goes multicore and adds a video processing system that allows the chips to tackle image processing chores in real time. The ADSP-BF60x family (Fig. 1) now has dual core version. The high end version also incorporates a Pipelined Vision Processor (PVP). The PVP (Fig. 2) allows the system to handle real time video processor chores would normally require an even higher performance processing sytems.
The ADSP-BF60x delivers hefty processing performance by itself. The top end BF609 has a pair of 500 MHz cores. Each has 148 Kbytes of L1 cache/memory with parity support. The shared 256 Kbyte L2 cache/memory has ECC support. This improved error checking support will allow the Blackfin to be used in safety critical applications where this is a requirement along with dual watchdog timers. Off-chip memory support includes DDR2 and LPDDR1. The Static Memory Controller (SMC) provides off-chip flash access. There is also an enhanced MMC (eMMC) flash controller.
The BF609 has an HD (1280 by 960 pixels) PVP while the BF608 has a PVP that handles VGA resolutions. The BF606 and BF607 have 256 Kbytes and 128 Kbytes of L2 cache respectively and do not include a PVP. These chips target applications that do not need real time (30 frames/s) video processing support.
The peripheral suite includes the usual communication peripherals such as I2C, SPI, CAN, SPORT, and serial ports. The four link ports are designed to connect multiple chips together. These high speed serial ports can also connect the DSP to FPGAs. The top end chips have a pair of 1588-compatible Ethernet MACs (EMAC).
The high speed ePPI (enhanced Parallel Peripheral Interfaces) ports can be linked to video streams such as digital cameras. The ePPO ports can be connected directly to the PVP through a crossbar pixel switch.
Pipelined Vision Processor consists of a dozen configurable blocks that can be tied together in a variety of ways. They are designed to work on a video stream and are sufficient to provide functions such as object detection, identification, and tracking. They keep track of a few lines of video but do not cache an entire frame. This reduces memory overhead resulting in lower power consumption. A typical application is traffic sign recognition (Fig. 3).
The PVP is designed to support the popular algorithms such as Sobel, Canny and Laplace. Analog Devices is also providing examples and configurations that developers can start with. The data flow can be from a camera or to a display via memory. The blocks can be mixed allowing dedicated or mixed data flows.
The chips initially target automotive applications where ADAS (advanced driver assistance systems) is becoming more common. Common functions include collision warning and mitigation, pedestrian detection, intelligent high beam control and traffic sign recognition. Lane departure warning and lane keep assist are also in the list.
Other applications include advanced barcode recognition and security camera analytics. Alternatives often require FPGA support to perform the same function. One of these Blackfin DSPs can be used to create a very intelligent network camera.
The latest Blackfin chips are going to make a significant impact in a range of applications from robotics to transportation. The PVP is a major addition providing flexibility that is more often found in FPGAs when it comes to video processing. The ability to support multiple video streams set the Blackfin apart from the competition.