Optical Product Quality As A Driver of Total Operating Costs

By Teh-Hua Ju on June 8, 2022 | Leave a Comment

High speed optical networks are always on, carrying traffic for data, video and personal communications. Availability is assumed, but it is determined by factors of product quality and operational reliability. This manifests in network operating expense for both direct operations and for maintenance or repair.

Optical module or components products with poor quality or reliability are much more expensive than their replacement cost alone! Added costs are associated with lost revenue for network down time, troubleshooting, repair or replacement, conducting root cause analysis, taking corrective actions, making possible design changes, and direct product recall costs. For networks carrying high speed traffic for communications, pursuing excellent product quality is essential to maintain suitable total costs of network ownership and operation.

Outages and downtime are expensive. A recent article in Forbesprovides some background and examples. The article reports that in a recent survey of service providers by the Uptime Institute, "over 60% of the respondents reported losing more than $100,000 to downtime. Of that 60%, and 15% lost over $1 million." The article goes on to point out that the impact of such outages goes far beyond the customers directly affected and can impact emergency services, financial services and company reputations.

Another article in Incident Managementdescribes concrete examples of the losses incurred, including a five-hour power outage in August 2016 that caused 2,000 cancelled flights and an estimated loss of $150 millionfor Delta Airlines. Or in March 2019, a 14-hour outage cost Facebook an estimated $90 million.

Data centers are a critical part of today's information infrastructure and they support connectivity and applications requiring networks that can scale, while delivering availability, security and reliability. Since the vast majority of communications within and between datacenters are carried out through fiber optic connections, such attributes depend directly on the optical products used in their networks.

Product quality starts from its inputs - robust product and process designs, well defined specifications, and disciplined manufacturing.

• Robust product/process designs - Product design is always coupled to the assembling/testing process and equipment used, and strong production yield is an indicator of the design quality, and simultaneously drives cost.
• Well defined specifications - from product specifications to process specs, then to material specs, all need to be defined clearly and completely integrated across all process methods.
• Disciplined manufacturing - POR 4M1E (Man, Machine, Material, Method, Environment) need to be tightly followed, deviations across manufacturing practices need to be reduced.

Manufacturers that always pursue continuous improvement, leveraging failure mode analysis in order to identify improvement opportunities and actions, will ultimately have the highest product quality, lifetime performance and the lowest total cost of ownership.

Products with high quality attributes per these metrics will provide high reliability and stable performance over the course of its service. It is instructive to look at one product as an example. In high-speed coherent networks, narrow linewidth tunable lasers are a high-volume product that are critical to network performance and, therefore, need to deliver high reliability and stable operation. Such lasers not only provide the light that carries the information in an optical fiber, but also provide a "local oscillator" which is critical for coherent systems to extract the data at the receiving end. Any degradation in performance in terms of output power, wavelength stability or the "purity" of the lasers "color" (i.e. the linewidth) will drastically degrade overall link performance. And, since in current fiber optic systems one channel or wavelength can carry 200G, 400G, or even 800 Gbits of information per second, the failure of a single laser can eliminate a lot of data capacity. Furthermore, lasers by their nature present reliability issues. The high optical power density at the laser mirrors and the high electrical power in the gain medium make them particularly challenging.

That is why such a laser should deliver less than 30FITs at 90% confidence level over its full operating lifetime and production volume. FIT stands for "Failure in Time" and is defined as the number of failures expected in one billion hours of device operation. Shown in the table below is a summary of the FIT data for such an ultra-narrow linewidth tunable laser. Over a period of approximately 7 years, 1.1 million such lasers have been shipped and accumulated 26 Billion total device hours of operation. Over that period, 579 failures have been reported. This would give a simple FIT rate of 579/25.865 = 22.385. However, a key word in the definition of FIT rate is "anticipated". Failures are a statistical process and a different batch of devices may have a different number of failures. Also shown in the table below are the FIT rates with various statistical confidence levels, ranging from 22.41 at 50% confidence to 23.63 at 90%. These FIT rates do not vary much due to the large sample size of 26 Billion operating hours. For newer products with less history, the determined FIT rate can have a much wider range.

Table 1: FIT Rate of Narrow Linewidth Tunable Laser (Micro-ITLA)

A FIT rate of 22 is very low, indicating a Mean Time to Failure of approximately 44 million hours, or 5000 years. So one might expect that a laser will never fail! But in a population of 1 million lasers, statistically, we would expect a failure every 44 hours. And when you consider that systems are comprised of hundreds or thousands of components, it is critical each one have the lowest possible FIT rate.