Chips JU 2024 IA NeAIxt

Purpose and goal

The NeAIxt consortium addresses challenges in next-generation Edge AI. The Swedish team develops energy-efficient custom solutions for object and situation recognition using IRnova’s infrared and mm-wave sensors, applied in health, safety, and industry with Strikersoft, FOI, and IRnova. Implementations run on KTH’s SiLago platform, ported to 22nm FDSOI for ASIC-like efficiency, benchmarked on COTS, and synthesized to IMEC’s 2nm node.

Expected effects and result

In NeAIxt, the Swedish consortium will deliver: FOI demonstrating neural-network object detection using IRnova’s advanced IR sensor; IRnova showcasing yield-improvement AI/ML and IR-based detection; Strikersoft enhancing search-and-rescue with mmWave and IR sensors; and KTH presenting its 22nm-ported SiLago framework for these demos, highlighting the potential benefits of a future 2nm port. KTH will also be able to show 10-100X improvement of SiLago designs to COTS implementation

Planned approach and implementation

Strikersoft, IRnova, and FOI will define quantifiable requirements for their AI/ML applications, prepare training and validation datasets, develop their algorithms, and implement and demonstrate them on COTS platforms such as FPGAs and GPUs. KTH will adapt the existing micro-architectural SiLago framework to the 22nm node, characterize and map partners’ AI/ML applications onto it, benchmark SiLago designs against COTS implementations, and synthesize SiLago for 2nm to evaluate potential benefits.