Monday, January 1, 2018

The Why and not the What of Receiver Test

It is important for an engineer in technical sales to be able to explain not just what "what" of test and measurement technology, but the "why".

I use this question in almost all of my interviews now for sales engineers, and it helps me see how well an engineer can explain a concept at a high level.  The conversation goes something like this...

Joel Avrunin: "Many Gen 1 and Gen 2 high speed serial standards only involve transmitter test (TX), requiring an oscilloscope.  Receiver test (RX) is limited or non-existent.  I am an engineering manager putting someone else's silicon into my design, and I am concerned about the cost of test.  So why do Gen 3 standards such as USB3.1 require RX test?"

Here is the wrong answer.

Applicant: "Because it is required by the compliance test."

Joel Avrunin: "But WHY does the compliance test require it?"

Applicant: "To make sure the eye diagram is open at the receiver."

Joel Avrunin: "So I'll use an oscilloscope like I've always done to make sure the eye is open!"

Applicant: "But you can't do that - they are closed eye standards"

Joel Avrunin: "Why are they closed eye standards?"

Applicant: "Because there is too much high frequency insertion loss in the dielectric of the cable, jitter from random noise sources, ISI, and you need to make sure your CTLE is......"

Joel Avrunin: "WHOAH!  Hold on... why can't I just use a better transmitter, higher quality cable, better interconnects, and get an open eye?"

Applicant: "Ummmmmm.... because it's a closed eye standard and the compliance test requires receiver test."

A sales engineer who works for me has to be able to explain the purpose of the test.  We are entering an era where more attention is paid to cost of test than ever.  When budgeting a new project, the cost of buying $1M of new test equipment (or even $100k) looms large over project budgets.  Astute managers see labs full of equipment that was "new" just 5 years ago - why isn't that good enough?  Especially outside the silicon world, receiver tests with products such as the Tektronix BERTScope just were not performed.

Everything the applicant said above was correct.  But nothing there explained why receiver test is required.

Here is the simple answer.

Cost of interconnect.

It drives everything.  Moving a high frequency signal 1mm across a piece of silicon is (relatively) simple.  Moving it from the silicon, through a package, into a PCB, through a via - complex but still (relatively) simple.  But get it to leave the safe world of the PCB and into a flexible cable and you start talking real cost to keep the signal intact.

This is a high quality, precision machined 2.92mm connector.  USB3.1 is a 10GB/s standard, and this is what test and measurement companies like Tektronix suggest you use to qualify your designs.

It has bandwidth to 40GHz+, must be screwed in and torqued with a precision calibrated torque wrench to exactly 8lb-inch (no more, no less).  The cable assembly itself has many layers to transmit the signal with little loss - most of these can cost in excess of $1000 and must be kept with rubber caps on the end to protect the threads of the connector.

We need that because fast data rates require fast edges from 0 to 1 and 1 to 0.  A fast edge is a combination of frequencies.  Fast edges with little jitter or ISI yield clear 1's and 0's (on the left below).  Remove the high frequency content, and the edge slows down.  Slow the edge down enough, and the eye "closes" and the data link fails to transmit data (on the right below).

This is a USB3.1 portable hard drive.  2TB of storage for $50 (as of the end of 2017), and it comes with a cable.  You will likely stick this in the pocket of your cargo pants along with dryer lint and half a Clif Bar you are saving for later.  Yet when you plug it into your computer, you expect to get 10GB/s of data transfer.

How do we pay $50 for the hard drive instead of $1000?

As you see below, the USB connector and cable assembly are not "lab grade".  You can see that instead of 1 shielded and controlled impedance cable with a machined precision connector that is torqued exactly, we have a bundle of wires (shielded, but nowhere near as controlled as the cable assembly above), with a connector that uses pins that pressure-fit onto a PCB in a shield that is stamped and wrapped around the connector.  It is a commercial quality connector, not a lab quality connector.

So how do we keep pushing data rates faster but keep the cost of our devices down?  Better silicon receivers - we keep the magic in the silicon!  That portable hard drive doesn't come with $1000 cable and a torque wrench.  It comes with a $5 cable that you plug in and the silicon inside handles all of the impairment.

Let's revisit the interview:

Joel Avrunin: "I am an engineering manager putting someone else's silicon into my design, and I am concerned about the cost of test.  So why do Gen 3 standards such as USB3.1 or PCIe require RX test?"

Applicant: "We want to move data rates faster than we did in the past, but we need to keep the cost of our interconnects down.  In some cases we have legacy backplanes or CEM connectors that can't be easily upgraded.  In other cases, we need consumer grade cable assemblies.  Receivers need to handle impaired signals, and a BERTScope Receiver test ensures that these faster standards can still be transmitted over lower quality interconnects".

What do you think?  Is there a better answer to this interview question?  Please let me know in the comments section below.

No comments:

Post a Comment