POLYN tapes out analogue neuromorphic chip

POLYN Technology recently achieved a significant milestone by completing the tape-out of its first product chip. This chip boasts an analogue neuromorphic core that is based on a Voice Activity Detection neural network model. The primary purpose of this chip is to validate the NASP (Neuromorphic Analogue Signal Processing) platform technology and the initial customer products. Additionally, it serves to showcase the performance and robustness of the NASP technology, marking a crucial step forward for POLYN.

According to Aleksandr Timofeev, the CEO of POLYN, this achievement represents a major advancement in the development of NASP chips. Timofeev stated, “This is a considerable milestone in NASP chips development toward our mission to bring energy-efficient neuromorphic solutions to market.” He emphasized the uniqueness of the NASP technology platform and highlighted the significant engineering expertise and intellectual property that have gone into its implementation.

POLYN has employed a sophisticated neural network compiler and a customized EDA flow using standard Cadence tools to convert a trained digital neural network model into analogue neuromorphic cores. This innovative approach enables the efficient creation of a cohesive neural network layout comprising thousands of neurons and connections with resistive weights. Importantly, this flow can be utilized at any semiconductor foundry, enhancing its versatility and applicability.

By seamlessly integrating Cadence Virtuoso for custom analogue design and Innovus Implementation System for digital place and route, clock tree synthesis, and timing closure with its proprietary NASP platform, POLYN has optimized the chip development process. This integration has enabled the engineering team to meet aggressive time-to-market goals and ensure the efficient realization of the neuromorphic chip.

The NASP technology developed by POLYN is at the forefront of edge and sensor signal AI processing, offering unique capabilities for cloudless, on-chip AI in various IoT devices. The POLYN neuromorphic architecture operates in a true parallel asynchronous mode, delivering exceptionally low latency and power consumption. Notably, the calculations performed by this technology do not rely on CPU usage or memory access, further enhancing its efficiency and performance across a range of real-time IoT applications.