The Raspberry Pi of AI: Korean startup aims to bring ultra-cheap, high-performance NPUs to billions of devices using its own secret quantification sauce, with a $1/TOPS target in sight
DeepX DeepX, Inc. is a South Korean AI technology company specializing in deep learning solutions in sectors such as autonomous systems, robotics, and healthcare. At the recent Embedded Vision Summit 2024, DeepX showcased its first-generation chips, the V1 and M1, designed for different applications, and hinted at its upcoming next-generation chip focused on AI for on-device and autonomous robotics applications.
The V1 SoC (formerly L1) features the DeepX 5-TOPS NPU paired with quad-RISC-V CPUs and a 12MP image signal processor. This sub-$10 SoC is built on Samsung’s 28nm process and runs the YOLO v7 model at 30fps while consuming only 1-2 watts. It supports the latest CNN algorithms for computer vision and is designed for products such as IP and CCTV cameras, robotic cameras, and drones.
The M1 is a larger accelerator designed to work with a host CPU. It is said to achieve the highest cost efficiency (inference/$), power efficiency (TOPS/W), and performance efficiency (FPS/TOPS). AI performance is 25-TOPS and it consumes 5W. It is suitable for use in consumer and industrial robots, machine visions, AI required IPC and HPC, smart factories, and edge computing.
Collaboration with LG
DeepX CEO Lokwon Kim told Sally Ward-Foxton of EE times that the company is working with LG to port LLMs to DeepX’s chip for use in mobile devices, cars and home appliances. “(In-device AI) really makes sense for their business model for LLMs, so that’s why we’re working together,” Kim said. “They’re providing their LLM technology so we can learn more about the characteristics of the model and optimize it for applications on the device.” The result is an NPU chip that’s optimized for running LLMs on the device, but initially it will only function as an accelerator. It’s expected to take another 3-5 years to develop a fully LLM-capable SoC.
The next chip on DeepX’s roadmap is the V3, designed in response to feedback from Chinese and Taiwanese customers. The V3 will reportedly feature a 15-TOPS dual-core DeepX NPU with quad-Arm Cortex A52 CPU cores and will operate at under 5 watts on average. “Previously we were using a RISC-V CPU, but customers wanted Arm,” Kim told Ward-Foxton. “So we looked at an Arm quad-core there. Customers also wanted USB 3.1, a more powerful ISP — not an upgrade on the NPU. So we redesigned it.”
If EE times explains: “Customers wanted Arm CPUs because the Arm ecosystem can provide better security solutions – many customers are building security camera systems. Other customers want to use the robot operating system, which is now supported on Arm, although it has not yet come out on RISC-V.”
DeepX says it will continue to offer the RISC-V-based V1 alongside the Arm-based V3 (samples of which are expected by the end of 2024), and promises to support both architectures going forward.