DATA CENTER

Layer 0 Copper Connectivity for the Future

Qing Xu
The industry is following the road to 800G Ethernet, as we’ve been discussing for a few months on the blog. As new Ethernet generations come into play, we’ve also been outlining the infrastructure and connectivity requirements that may be needed as a result.

Next-generation 200G and 400G Ethernet interconnects will still be supported by copper and fiber connectivity. Here, we’re outlining the requirements that data centers will face, and explaining why layer 0 (cabling and connectivity infrastructure) is becoming increasingly critical in supporting ultra-high data rates.

 

Differential Twinaxial Cable and Direct Attach Copper Assemblies

Twinaxial (twinax) copper cable is constructed with differential copper pair with ~100Ω impedance, and a drain (ground) wire for differential high-speed signaling.

  • In a typical one-lane direct attach copper (DAC) assembly (e.g. SFP+), two pairs of twinaxial cable are bundled together in the same jacket.
  • In a four-lane DAC assembly (e.g. QSFP+), eight pairs of twinaxial cable are bundled together in the same jacket.

One-Lane-vs-Four-Lane
Twinaxial copper cable construction: one-lane DAC and four-lane DAC

 

Today, twinaxial-based passive DAC is the de facto intra-rack interconnect solution (server to ToR switch) for most cloud data centers, mainly thanks to its low cost and solid performance in short-reach applications.

  • The main characteristic of merit – insertion loss of the differential twinaxial pair (SDD21) – is measured at the Nyquist frequency (half the signaling rate).
  • The frequency range of interest (cable suck-out free range frequency) is typically 0.75 × signaling rate.
 

 

As data rates increase, however, the maximum reach of DAC shrinks. For each new generation, there is roughly 30% less reach as a result of much higher insertion loss at the Nyquist frequency.

 

High-Speed-Twinax
Belden high-speed twinaxial cable insertion loss (5 m for 25GBASE-CR and 50GBASE-CR)

 

Compared to simple NRZ (non-return to zero) modulation, however, PAM4 modulation allows data rates to double without replacing the DAC cable assembly – but it comes at a cost of reduced signal-to-noise ratio, requiring more complex digital signal processing at the host (active system).

 

Maximum-Reach
DAC maximum reach vs. transmission speed (per lane)

 

On-Board Copper Cable Assemblies

As data lane rates move from 25G to 50G PAM4, and then to 100G PAM4, the chip-to-module (OIF’s CEI-56G-VSR and OIF’s CEI-112G-VSR) interface becomes very challenging to meet maximum link loss requirements on the traditional backplane – even with premium low-loss PCB materials like MEGTRON 7.

 

Twinaxial cable, on the other hand, exhibits much lower loss over short distances (up to 35 cm for VSR [very short reach], 50 cm for MR [middle reach] and 1 m for LR [long reach]).

 

A new application using internal twinaxial cable to replace the lossy copper backplane can largely improve signal integrity for high data rates, such as 50G and 100G data transmission in a PAM4 format. Several high-speed cable assembly vendors have already developed workable solutions, raising industry interest in overcoming the signal integrity bottleneck.

 

Traditional-PCB-vs-Twinax 

As we look down the road to 800G, make sure to subscribe to our blog updates so you can follow along with us. We’ll be talking about:

  • Layer 0 fiber connectivity
  • The ubiquity of 100G Ethernet
  • Optical fiber cabling migration toward 100G, 200G, 400G and 800G Ethernet

Our 800G Ethernet blog series has already covered:

Traditional PCB vs. twinaxial on-board copper assembly

 

Belden’s end-to-end copper and fiber cabling and connectivity will help you prepare for 800G Ethernet – and every stop in between. Learn more here.