Transceiver design a key for 16G Fibre Channel success

By Raju Kankipati

Overview

With 16-Gbps Fibre Channel now in the standardization process, transceiver vendors have already begun work on appropriate modules. But the expected requirements pose numerous challenges.

Fibre Channel dominates the storage area network (SAN) and external storage marketplace, providing high-speed connectivity between computing and storage resources. It is also a reliable and cost-effective approach for other types of networks that support video, data acquisition, and many other applications.

The T11.2 Technical Committee of the International Committee for Information Technology Standards (INCITS), an American National Standards Institute (ANSI)-accredited standards committee, has established Fibre Channel’s specification standards. Fibre Channel data rates start at 1 Gbps and scale up to 2, 4, and 8 Gbps. However, applications such as video and data backup are driving a demand for even higher data rates. To meet customers’ evolving needs for faster speeds and lower cost, the committee is standardizing Fibre Channel Physical Interface-5 (FC-PI-5), the next-generation data rate for Fibre Channel. This standard, commonly referred to as 16G Fibre Channel, aims to double the throughput of the 8G standard with a defined line rate of 14.025 Gbps.

The majority of today’s Fibre Channel deployments use 4-Gbps connections; 8-Gbps transceivers in the SFP+ form factor are just starting to ship in small volumes. Meanwhile, Fibre Channel over Ethernet (FCoE)—another Fibre Channel standard intended to enable LAN and SAN convergence in data centers to reduce network complexity—maps Fibre Channel frames into 10-Gigabit Ethernet frames without compromising the low-latency, lossless characteristics of Fibre Channel. FCoE is gaining momentum in the industry and is supported by a lot of big storage system vendors, but its success in the market place has yet to be determined.

Nevertheless, work has begun on 16G Fibre Channel transceivers, and their form is beginning to take shape. The basic assumption is that such modules will use the SFP+ form factor and, when the market matures, the cost will be similar to 8-Gbps transceivers. (Such maturation usually takes about two to three years after the standard is established.) The data rate (14.025 Gbps) and encoding scheme (64b/66b) have been finalized, and other specifications might be completed by end of 2009. The two port types defined by the standard are 16-Gbps shortwave (SW) for multimode-fiber applications and 16-Gbps longwave (LW) for singlemode-fiber applications.

Design challenges

The preferred form factor and pin-out for 16-Gbps Fibre Channel transceivers is SFP+ (see photo). The target distance for multimode-fiber applications SFP+ is 100 to 150 m and 10 km for singlemode-fiber applications. Like all previous Fibre Channel protocols, the Fibre Channel Industry Association (FCIA) requests suppliers to support two generations backward compatibility. Backward compatibility enables SAN switches to be upgraded to a higher capacity when the infrastructure is ready and ensures investment protection. Thus, 16G transceivers have to be compliant to 8G and 4G specifications to meet this need.

All of these requirements pose a few challenges to transceiver designers. SFP+ transceivers typically have a laser driver and a transmitter optical subassembly (TOSA) to transmit data via fiber (see Fig. 1). On the Rx side, they have a receiver optical subassembly (ROSA) with a built-in transimpedance amplifier (TIA) and potentially a post-amplifier for limiting interfaces. For 16G SFP+ transceivers, a clock data recovery (CDR) IC might be required for robust performance. Built-in CDRs help clean the jitter before sending the signal to the laser. A microcontroller is also required inside the SFP+ to implement the software management interface.

Fitting these additional components inside an SFP+ package and still staying within the 1-W power budget will be a challenge with the existing components. The use of 65-nm complementary metal-oxide (CMOS) process technology for ICs will help to reduce size and power consumption. Some IC integration efforts are ongoing in the industry in order to develop a one-chip option for CDR/laser driver and CDR/post-amplifier.

For multimode-fiber applications, CDRs are required both on the transmitter side and the receiver side. To use existing 10-Gbps vertical cavity surface-emitting lasers (VCSELs) for 16G SFP+, the reliability and yield might have to be improved to meet the spectral width and rise/fall time specification. For singlemode-fiber applications, the use of distributed feedback (DFB) lasers can eliminate the need for a CDR on the transmit side. Figure 2 shows an eye diagram of the first 16G SFP+ LW prototype using a DFB laser. Uncooled, directly modulated DFB lasers can also be used for lower cost and lower power consumption.

There might be some challenges when operating 16G SFP+ at 4 Gbps to interoperate with a 4G transceiver on the other end of the link. The 16G SFP+ might see overshoot and undershoot of the signals coming out of the 4G transceivers. This challenge can be overcome if the host ASIC can send a signal to the 16G SFP+ transceiver indicating the signal speed instead of autonegotiation. With this signal from the host, the transceiver can adjust the receiver for error-free operation at 4 Gbps.

On the system ASIC side, some transmit pre-emphasis and receive equalization for the electrical signals might be needed to enable reasonable PCB trace lengths. This requirement is similar to 10G SFP+ transceivers today. Pre-emphasis on the host ASIC in combination with the CDR built into the SFP+ transceiver will help mitigate signal degradation due to PCB and external media impairments as the data is transferred and converted into an optical signal. This will also improve the jitter and the optical eye mask margin of the transmitted data. An equalizer, like an electronic dispersion compensator (EDC), on the ASIC receiver side will compensate for the signal degradation on that side.

A few transceiver companies have already demonstrated 16G SFP+ SW and LW. Based on these demonstrations, the availability of integrated chipsets, and the completion of standards specification, first prototypes of 16G SFP+ SW and LW for customer evaluations are on the horizon; shipments in small quantities might start in late 2010.

For a 16G-based SAN to run optimally, all the core infrastructure pieces such as the host board adapters (HBAs), servers, switches, and storage systems have to be upgraded to 16G Fibre Channel. Some sort of interoperability effort, similar to 10G and FCoE SFP+, has to occur for 16G SFP+ transceivers in these core network pieces before market adoption and actual deployment.

Meanwhile, the next logical step for Fibre Channel after 16G will be 32G. But this step will depend on actual market demand. Figure 3 shows DFB laser technology for 32G longwave applications.

Conclusion

SFP+ transceivers for 8G Fibre Channel have just started shipping in volume and are being deployed. There is a lot of interest in the market for FCoE to drive convergence in the enterprise data center. The T11.2 task force of INCITS standards committee is specifying the next-generation Fibre Channel standard FC-PI-5 for 16G. Early prototypes of 16G SFP+ are on the horizon, but actual transition from 8G to 16G deployment will depend on how quickly the standard is established and how system vendors work with optics suppliers to overcome the technical challenges of performance and power consumption.

Adoption will also depend on cost and when 16G SFP+ will be available for only a small price premium over 8G SFP+. The logical step for Fibre Channel after 16G will be 32G, but the leap from 16G to higher speeds will depend on actual market demand.