Murata — yes, the capacitor company — just dropped a co-packaged optics roadmap through 2032.
The play: mount the optical engine directly next to the switch chip. Shorter copper runs. Light does the heavy lifting. Lower loss, lower power.
Three core pieces:
1. Thin-film lithium niobate modulator. Repurposed from RF filter tech. Target: >100 GHz bandwidth, 400 Gb/s per lane by sample in Q2. Mass production 2028. Bump to 800 Gb/s per lane by 2030.
2. Ceramic substrate. Better thermal and moisture handling than organic. 125mm sample already runs 400G electrical. 2028: embedded cooling and power. 2030: on-board optical routing, 200mm panel.
3. Flexible plastic interconnect. Bends in tight spaces, carries 400G electrical today. Optical version by 2030. Unified board that switches between electrical and optical based on distance: 2032.
The tradeoff: repairability. When optics and ASIC share a package, you can't hot-swap a failed module. Murata is still working the spec with hyperscalers.
This is the supply chain catching up to the bandwidth wall. If it works, it's a new category. If it doesn't, it's an expensive science project.
The play: mount the optical engine directly next to the switch chip. Shorter copper runs. Light does the heavy lifting. Lower loss, lower power.
Three core pieces:
1. Thin-film lithium niobate modulator. Repurposed from RF filter tech. Target: >100 GHz bandwidth, 400 Gb/s per lane by sample in Q2. Mass production 2028. Bump to 800 Gb/s per lane by 2030.
2. Ceramic substrate. Better thermal and moisture handling than organic. 125mm sample already runs 400G electrical. 2028: embedded cooling and power. 2030: on-board optical routing, 200mm panel.
3. Flexible plastic interconnect. Bends in tight spaces, carries 400G electrical today. Optical version by 2030. Unified board that switches between electrical and optical based on distance: 2032.
The tradeoff: repairability. When optics and ASIC share a package, you can't hot-swap a failed module. Murata is still working the spec with hyperscalers.
This is the supply chain catching up to the bandwidth wall. If it works, it's a new category. If it doesn't, it's an expensive science project.
