Key Takeaways

• Nokia is running Cadence’s AuraStack AI Super Agent on the Millennium M2000 AI supercomputer, delivering a 5× acceleration of core design and analysis tasks.
• The solution fuses agentic AI, GPU‑optimized EDA software, and scalable accelerated hardware in a fully on‑premises, data‑secure environment.
• AuraStack automates schematic capture, placement, routing and multi‑objective optimization using NVIDIA Nemotron open‑model AI.
• Cadence’s computational stack leverages CUDA‑X, cuDSS, and NVIDIA Omniverse to speed power‑integrity, thermal, EM and circuit‑simulation flows.
• The modular Millennium architecture lets Nokia expand compute capacity and run multiple AI‑driven workloads in parallel, establishing a new template for large‑scale electronic‑system design.

Cadence‑Nokia Partnership Powers AI‑First Electronic System Design

Why the Move to AI Matters

Designing modern heterogeneous electronic systems—5G radios, automotive ASICs, IoT edge modules—requires simultaneous trade‑offs across power, timing, thermal and electromagnetic domains. Traditional manual or rule‑based flows struggle to explore the combinatorial design space within realistic project timelines. By embedding agentic AI directly into the design kernel, Cadence enables Nokia to evaluate thousands of alternatives in minutes rather than weeks, shortening time‑to‑market while safeguarding proprietary data behind an on‑premises fence.

Architecture Overview

1. Agentic AI – AuraStack Super Agent

• Core model: NVIDIA Nemotron open‑model foundation, fine‑tuned on Cadence’s electronic‑design datasets.
• Functions automated: schematic capture, component placement, detailed routing, multi‑objective (power‑delay‑area) optimization.
• Outcome: 5× faster completion of typical “place‑and‑route” runs (e.g., a 12‑hour manual flow reduced to ~2.4 h).

2. Computational Software – GPU‑Optimized EDA Stack

• Performance metric: Benchmarks show up to 8× reduction in simulation wall‑time for 10 M‑gate mixed‑signal blocks versus CPU‑only runs.

3. Accelerated Computing – Millennium M2000 Platform

• Compute fabric: 4 × NVIDIA H100 Tensor‑Core GPUs (each 60 TFLOPS FP64, 1 PFLOPS FP16).
• Memory: 1.5 TB high‑bandwidth DDR5 + 256 GB HBM2e per GPU.
• Scalability: Modular chassis allow incremental addition of GPU nodes; capacity can grow from 4 to 16 GPUs without re‑architecting the software stack.

How Nokia Leverages the Stack

Parallel AI‑Driven Workloads

Nokia runs multiple AuraStack instances concurrently—one for RF front‑end layout, another for power‑delivery network (PDN) optimization, and a third for thermal‑aware floorplanning. The shared M2000 fabric schedules these jobs dynamically, ensuring maximal GPU utilization (> 85 % average) and eliminating idle compute windows.

Secure, On‑Premises Deployment

All AI models, design files and simulation results remain inside Nokia’s data center, complying with ISO 27001 and GDPR requirements. The platform uses encrypted inter‑node communication and role‑based access controls, eliminating the risk associated with cloud‑only AI services.

Comparative Snapshot: AI‑Enhanced vs. Conventional Flow

All figures are based on Nokia’s internal benchmark suite for a 7 nm RF‑SoC design.

Bottom Line

By coupling Cadence’s AuraStack AI Super Agent with the Millennium M2000 GPU supercomputer, Nokia has built a self‑contained, AI‑first design environment that slashes critical workflow times by up to 5× while preserving strict data security. The three‑layer stack—agentic AI, GPU‑accelerated EDA software, and modular high‑performance hardware—creates a scalable foundation that other systems‑level companies can emulate to meet the escalating complexity of next‑generation electronic products.