AIStorm addresses the enormous opportunity of enabling more 'intelligence' right at the edge of the network.
1. Company and team
- This low-profile AI tech startup is poised to tackle a big problem: enabling faster processing of complex AI problems at the very edge of the network — within sensors.
- AIStorm has developed and patented a new approach that will disrupt the GPU-based approach that is typically used today.
- Founded/led by David Schie, an expert in analog and mixed-signal hardware design who has led large teams at Maxim, Micrel, and Semtech.
- The team also includes proven veterans that round out the company's design and fabrication expertise.
2. Market & technology
- AIStorm has solved a growing problem: the need to process sensor information at the edge of the network to reduce the cost and security risk of transmitting large amounts of raw data.
- AI systems require information be available in digital form before they can process data, but sensor data is generally analog. AIStorm solves this problem by processing sensor data directly in its native analog form, in real time.
- AIStorm is targeting some of the world’s largest handset, machine vision, wearable, IoT, automotive, food service, AI assistant, security, biometric device, & imager applications.
- Gryfalcon, Mythic-AI, Syntiant and Google have all announced they are pursuing AI engine platforms at the edge. But... none of their approaches incorporate sensors into the solution. AIStorm changes that.
AIStorm is backed by four large sensor and industrial companies that are eager to integrate AIStorm's technology into upcoming products…
- Egis Technology, a major biometrics supplier to handsets, gaming, and advanced driver-assistance systems.
- TowerJazz, the global specialty foundry leader that specializes in image sensors for commercial, industrial, AR, and medical markets.
- Meyer Corporation, a world leader in food preparation equipment.
- Linear Dimensions Semiconductor, a leader in biometric authentication and digital health products.
David Schie's US patent application 20140344200 "Low power integrated analog mathematical engine" shows a convolution engine based on partial charging of caps by the controlled current sources, said to eliminate the need in capacitor scaling to change coefficients:
"The switched capacitor charge controls allow for nodal control of charge transfer based switched capacitor circuits. The method reduces reliance on passive component programmable arrays to produce programmable switched capacitor circuit coefficients. The switched capacitor circuits are dynamically scaled without having to rely on unit passives, such as unit capacitors, and the complexities of switching these capacitors into and out of circuit. The current, and thus the charge transferred is controlled at a nodal level, and the current rather than the capacitors are scaled providing a more accurate result in addition to saving silicon area. Furthermore, the weightings and biases now set as currents may be saved and recalled by coupling current source bias circuits to non-volatile memory means such as analog non-volatile memory."