PhD Intern, AI ML in Wireless L1/L2 - Fall 2026

Role Overview

NVIDIA Aerial CUDA Accelerated RAN (ACAR) is a framework for building high-performance, software-defined, cloud-native Radio Access Network functions over NVIDIA CPU/GPU/DPU based systems. We are seeking a self-motivated Intern to drive the adoption of AI/ML functions in the Phy and Mac layers of our Aerial SW, advancing the field of AI native wireless stack to achieve the Spectral and Energy efficiency goals of 6G.

Responsibilities

• Develop and Optimize AI/ML modules for functional blocks specifically in wireless signal processing
• Perform literature survey to understand the prior art on AI/ML for RAN
• Analyze and identify the suitable ML architecture for the RAN functions of interest
• Identify the right ML Architecture, complexity for each of the functional blocks
• Collaborate with multi-functional teams to optimize the OTA performance and compute complexity
• Benchmarking of OTA performance improvements with AI models and compute needs on different platforms
• Iteratively train, test & modify Model Arch for performance improvements

Requirements

• Full time PhD student doing research in the fields of AI and Wireless domains
• Thorough understanding of the wireless Layer1/Layer2 functions and algorithm aspects
• Excellent grip on AI and ML concepts, techniques and abreast of latest developments
• Deep understanding of Transformers, CNNs and other ML Architectures
• Hands on experience in simulating signal processing algorithms in Matlab and Python
• Programming skills in C/C++
• Experience in analyzing the problem, identifying the right model architectures, developing Models, Training and Optimization

Skills

• Python
• C/C++
• Matlab
• CUDA
• Machine Learning

Nice to Have

• Knowledge of CPU, DSP or GPU architecture, as well as memory, I/O and networking interfaces
• Experience with programming latency sensitive, real-time, multi-threaded applications on CPUs and one or more of GPUs or DSPs or Vector processors
• Familiarity with CUDA programming and NVIDIA GPU Architectures