Monday, December 28, 2009

Learning Lessons

We are finally shipping our RJ transceivers. Just about the size of an RJ45 copper jack, these transceivers are roughly 1/3 the size of existing SFF or SFP modules which is a benefit to the embedded marketplace where board space comes at a premium. They feature a screw down/surface mount solder attachment process that works much better in rugged environments than pluggable SFPs. And, we have a metal clip inside that secures the laser and receiver to the case and board to minimize damage from cable/connector vibration. We’ve been in development on these parts for over a year and now have an SX specification part, 850nm on multimode, in production. Our first big shipments happen before the end of the year with two versions available, 100Mb to 2Gbps, and 2.5G to 4Gpbs speeds.

We designed our own cast metal case, sheet metal top, screw mount to secure the OSAs, connector, circuit board inside with a very unique flex circuit, and surface mount connector…all done from scratch. The RJ parts also feature our own transceiver controller, called Leroy, which does all the magical things to control transceivers over temperature that the big ASICS inside of Finisar and Avago also do. Over temperature, the optical output values move less than 2 dB output power, is almost immovable on Rx Sense, and varies little on Extinction Ratio and OMA. Leroy also gets us Digital Diagnostics per SFF-8472.

We have a competitor that makes a part similar to our RJ but lacks digital diagnostics and a controller like Leroy. With the controller, we can select optical performance metrics based upon speed or link budget. The changes to the transceiver for these operating parameters are in the laser where there two variables:
• Modulation current, which in general, controls the height of the optical eye (OMA) as the light power goes on and off. The modulation is at the same rate (speed) as the data.
• Bias current, in general, controls the average power of the laser or its output power.

All our lasers are packaged with a monitor diode, a small receiver in the laser package. The monitor diode has a ‘slow’ response time – it can’t see the ‘fast’ modulation rate. It can only see the average power. Therefore, it can only help to control the Bias current. To control the modulation current, we use a curve fit or a table depending upon the part. The curve is simple and straightforward but requires a lot of work up front to qualify lasers, boards, assembled parts, and test parameters. We ship what we think will work, but don’t have any feedback to know its working. Any part without Digital Diagnostics operates without feedback, but at least our parts have a curve fit to match shipped and tested parts. For some applications, especially as speed increases or temperature range widens, we employ a table with feedback. This keeps the optical performance in specification, i.e. Ethernet, Fibre Channel, etc, over time and temperature. If a potential failure or failure occurs, the transceiver can let the board it’s on know. The best means of controlling the transceiver and laser in particular, is to use the extra pins we offer to connect via an i2c bus and set the speed or parameters for the specific application.

COTSWORKS transceivers are designed to last a long time and we back that up with support to help meet military, aerospace, or industrial specifications. In the last four years we’ve shipped nearly 50,000 transceivers and seen what happens to Commercial Off The Shelf transceivers in harsh environments. We learned lessons by trial and error…and trial again. While we’ve been in development on the RJs for only one year, there are really four years of experience at work.

Something to remember about learning lessons: the day you make the mistake is rarely the day you learn the lesson from it. With the RJs, we’re trying to learn from mistakes and invention over our whole time in business. Another learned lesson? The shortest path between two points in innovation and invention is not a straight line. And traveling the crooked road takes more time than what’s put in the specific project schedule.

No comments:

Post a Comment