The Laceli AI Compute Stick developed by Gyrfalcon Technology is directly competing with the deep learning Neural Compute Stick lanuched by INtel last year after its acqusition of chip designer Movidius.
The Gyrfalcon design uses its Lightspeeur 2801S AI processor that provides 2.8 TOPS (2800M operation per secod) with just 300mW of power, which the company says makes it 90 times more efficient than the Movidius USB Stick at 0.1 TOPS within 1W.
Lightspeeur is the first instantiation of Gyrfalcon's APiM architecture, which uses memory as the AI processing unit. This eliminates the huge data movement that results in high power consumption. The architecture features true, on-chip parallelism, in situ computing, and eliminates memory bottlenecks. It has roughly 28,000 parallel computing cores and does not require external memory for AI inference.
The standard USB 3.0 Stick can be connected to a range of devices such as PCs, laptops and mobile phones to enhance their image-based deep learning in applications such as image and video recognition, natural speech understanding, natural language processing, and day-to-day AI applications.
The Laceli AI Compute Stick will be shown in public for the first time later this month at CES 2018, although it is already in production.