Home  /  Solutions  /  Semiconductor Industry  /  Technical Guide
Technical Guide

Lithography Data Transfer: How 2×100G Fiber NICs with RDMA Keep Mask Data Flowing

In semiconductor manufacturing, the lithography machine writes circuit patterns onto wafers — but only if it receives the mask layout data in time. A 2×100G fiber NIC with RDMA builds the high-speed, low-latency channel this process depends on.

Why lithography demands high-speed data transfer

A single mask layout (reticle) file for a modern process node can reach tens of gigabytes. During production, this data must travel continuously between the lithography machine, storage servers and AI servers for OPC (optical proximity correction) processing and recipe management. Any stall in this pipeline means the stepper idles — and in a fab, idle time is measured in lost wafers per hour.

The network topology: stepper, storage, AI

The typical setup connects the lithography machine to storage servers holding reticle libraries, and to AI servers running pattern verification. Data flows in both directions and must be delivered with minimal jitter. A dual-port 100G fiber NIC in the storage or AI server provides the aggregated bandwidth and redundancy this architecture needs, connecting to the stepper through switches or directly to FPGA boards.

Why fiber, and why RDMA

Copper links struggle beyond a few meters with signal integrity and electromagnetic interference — a real concern on a fab floor full of high-power equipment. Fiber preserves signal quality over workshop and cross-plant distances. RDMA (Remote Direct Memory Access) lets the NIC move data straight into application memory, bypassing the CPU and OS stack. The result: dramatically lower latency and CPU load, and predictable transfer times even under load.

Tier options from 25G to 200G

Not every process needs 200G today, but the fab that standardizes on a tier can scale without re-architecting. LR-LINK covers the full ladder: 25G dual-port fiber NICs for edge and tool-side connections, 100G dual-port RDMA smart NICs for the storage backbone, and 200G dual-port RDMA NICs for the AI compute cluster. Controllers include Intel E810 and Broadcom options to match existing server ecosystems.

Recommended Products

Related Guides

Need help sizing lithography data transfer for your fab? Request a free trial of our fiber NICs and validate on your own test bench.

Request Free Trial