The Paradigm Shift in High-Speed Data Rates
As cloud data centers, AI accelerators, and telecommunications backbones scale to handle massive data throughput, SerDes architectures have transitioned from traditional 10Gbps and 28Gbps non-return-to-zero (NRZ) links up to 56Gbps and 112Gbps per channel. At these extreme frequencies, physical phenomena that were once minor layout nuisances become catastrophic signal integrity bottlenecks.
At 112Gbps, fundamental Nyquist frequencies reach 28 GHz to 56 GHz, where the signal wavelength in standard PCB dielectrics shrinks to a few millimeters. At this scale, every mechanical tolerance inside a connector behaves as a major transmission line discontinuity.
PAM4 Modulation: Doubling Bandwidth at the Cost of SNR
To achieve 56Gbps and 112Gbps throughputs without requiring impossibly high clock frequencies, the networking industry replaced binary NRZ signaling with 4-Level Pulse Amplitude Modulation (PAM4).
- NRZ (2 Levels): Transmits 1 bit per symbol cycle using two voltage levels (Logic 0 and 1).
- PAM4 (4 Levels): Transmits 2 bits per symbol cycle using four distinct voltage levels (00, 01, 10, 11).
While PAM4 halves the required fundamental Nyquist frequency for a given bit rate, it splits the eye diagram vertically into three smaller eyes. This reduces the available signal amplitude (eye height) by 9.54 dB, dramatically shrinking the system noise margin. Consequently, connectors in PAM4 systems must exhibit exceptionally low crosstalk, minimal return loss, and virtually zero impedance variation.
Physical Challenges at 56G/112G Frequencies
Designing connectors capable of operating at 112Gbps PAM4 requires overcoming several key physical loss mechanisms:
Severe Skin Effect and Surface Roughness
At 28 GHz, electrical current flows entirely within a tiny "skin depth" (less than 0.5 micrometers) on the outer surface of metal contact pins. Microscopic copper surface roughness, which is imperceptible at lower frequencies, acts as a long, highly resistive path, causing severe attenuation (S21 loss) and phase distortion across the contact interface.
Dielectric Loss Tangent
As signal frequencies pass into the microwave spectrum, energy is absorbed directly by the plastic polymers supporting the connector contacts. Connectors running at 112G require specialized low-loss Liquid Crystal Polymers (LCP) with extremely stable dielectric constants and minimal loss tangents across broad temperature ranges.
Micro-Reflections and Physical Stubs
At 112Gbps, physical features as small as a 0.2mm press-fit pin stub or an uncompensated solder fillet introduce quarter-wavelength resonances. These micro-reflections create ripple in the frequency domain that degrades the PAM4 eye opening.
Advanced Architectural Solutions for 112G Interconnects
To reliably support 112Gbps SerDes channels, connector manufacturers and systems architects employ radical design innovations:
- Overcoming Board Loss via Near-Chip and Flyover Cables: At 28 GHz, transmitting signals across several inches of standard FR4 or ultra-low-loss PCB substrate consumes the entire channel loss budget. Direct-attach flyover cable architectures plug high-speed twinaxial cables directly adjacent to the ASIC package, bypassing the PCB entirely and routing signals through air to the I/O panel.
- Precision Stamped Contact Leads: Modern 112G backplane connectors replace traditional pin-and-socket configurations with ultra-short, precision-stamped lead frames featuring broadside-coupled differential pairs enclosed in individual metallic Faraday cages.
- Eliminating Glass-Weave Skew: At 112G rates, slight differences in glass thread weave inside the PCB laminate cause intra-pair skew. Connectors must be designed alongside specialized spread-glass PCB substrates to ensure balanced phase arrival at the receiver.