Movidius Beefs up HW Acceleration in AI Chip

　　Myriad 2, an always-on many-core vision-processing unit, has already snagged big design wins from drone and surveillance companies. Working with customers and partners, El-Ouazzane explained, Movidius “has gained real-world experience,” which has taught the company acceleration needs in hardware blocks.

　　But with Myriad 2, “We were accelerating a lot of neural network workload in software,” he added. In contrast, Myriad X — described by Movidius as “a dedicated neural compute engine” — comes with a lot more microarchitecture in its hardware to accelerate deep-learning inference.

　　With many more hardware acceleration blocks, Myriad X architecture can do 1 trillion operations per second (TOPS) of compute performance on deep-neural network inferences, said El-Ouazzane. “And we keep it within a watt.” This is “an order of magnitude” faster than Myriad 2, he added.

　　So, what’s inside Myriad X?

　　Then, Movidius added a neural compute engine consisting of more than 20 enhanced hardware accelerators.

　　Noting that deep learning is “a fast-moving world,” El-Ouazzne said, “Your architecture needs to deal with new types of workloads in hardware microarchitecture, while DSPs are super useful in running new types of computer vision and deep learning algorithms.”

　　Myriad X offers increased configurable MIPI lanes. It connects up to 8 HD resolution RGB cameras directly to Myriad X with 16 MIPI lanes included in a rich set of interfaces, to support up to 700 million pixels per second of image signal processing throughput.

　　Beyond hardware accelerators, significant to keeping the power budget to 1 watt is memory architecture. Minimizing off-chip data transfer reduces latency and power consumption.  In Myriad X, the team increased on-chip memory to 2.5 megabytes, compared to 2MB in Myriad 2.

　　To carry neural net weights, external DRAM is necessary. Myriad X, now using Low Power Double Data Rate (LPDDR) 4 instead of LPDDR 3 in Myriad 3, incorporates 4 gigabits of DRAM in an 8 x 8.8mm package.

　　Asked why Movidius is not using Intel’s fab for Myriad X, El-Ouazzane said, “This was already in development well before Intel acquired us.”

　　Complex pipelines

　　To meet the challenge, he said, “You need dedicated hardware blocks.”

　　Movidius is exploiting what the team has learned from customers with Myriad 2 — for designing Myriad X that combines imaging, visual processing and deep-learning inference in real time.