Archive for January, 2011

The Meaning of IPC Cleanliness Designators

Friday, January 28th, 2011

An IPC cleanliness designator (defined in IPC J-STD-001) is sometimes called out on assembly drawings.  A common designator is C-22.

The first number indicates the number of board sides that must be cleaned
0 – No cleaning required
1 – Solder side only must be cleaned
2 – Both sides of the board must be cleaned.

The second and following numbers define the types of cleanliness tests to be performed:
0 – No test for cleanliness required
1 – Test for rosin residues – Used when high solids rosin fluxes are used in assembly.
2 – Test for ionic residues – This is a basic inexpensive test to detect insufficient cleaning.
3 – Test surface insulation resistance (SIR) – This test exposes the assembly to high temperature and humidity to accelerate failure mechanism that lead to shorts.
4 – Test for other organic contaminants – Type and amount of residue is measured, usually using a spectrometer.  This is an expensive test.
5 – Other tests requested by client – Any other testing beyond tests #1 – #4.

It is permitted to call out any combination of test types.  If both sides are to be clean and all tests to be performed, the cleanliness code would be C-212345.

Engineering Mindset Affects View of Precision

Wednesday, January 19th, 2011

Part of learning engineering is learning what requires precision and learning to let go and be imprecise when necessary.

I recall being in Dr. O’Malley’s circuits class at the University of Florida in the mid 90s. He drew an op-amp on the board and said it’s a very high-gain differential voltage amplifier. Someone asked what gain we should use. He said, “You can use whatever big number you want. For this example let’s use, I don’t know, say a million.” Several people laughed thinking it was a joke. O’Malley didn’t appear to understand the laughter. He continued, “You could use a similar number like two hundred thousand or five million. Anyway, that’s not the point. I hope it’s obvious that…”

I was amazed that 200,000 and 5,000,000 were very similar for the purposes of understanding an op-amp, so amazed that I remember it 15 years later. The statement seems perfectly natural now. Engineering has changed my mindset on precision.

This mindset affects aspects of life completely outside of engineering. The other night my wife, who is an attorney, said our baby was 10 minutes away from needing another dose of Tylenol for teething. “10 minutes?” I asked, “The tolerances can’t be that tight. That’s only a few percent of the dose duration.”

“It says 80mg every four hours” she said matter-of-factly, as if quoting the Uniform Commercial Code.

“But it doesn’t give a tolerance. I bet if you plotted the half-life of Tylenol in population of healthy babies, you’d find a distribution with half-lives varying by more than an hour. Even accounting for that, different babies will experience a therapeutic response at different blood concentrations.”

She said she felt more comfortable with a very clearly defined rule. Those tolerances made her uneasy. I felt more comfortable with a system that does not depend on any element having very tight tolerances. I wonder to what extent similar professional mindsets affect the daily lives of people in other fields such as accounting, medicine, science, teaching, and so on.

Quasi-Bidirectional Ports Can Cause Confusion

Monday, January 10th, 2011

I had another run-in with “quasi-bidirectional” ports today.  As on the last project where I encountered them, they caused confusion.  Today’s encounter was with the PCA9501 I/O port chip.

They should be called “open-drain outputs with moderate pulls connected to an input”.

They are open-drain outputs and can only be used as input ports by writing a logic high and then reading.

The thing that makes them tricky is the built-in pullup in moderately strong.  In the circuit I encountered today uses a quasi-bidirectional port as an output to drive the base of a ‘3904.  The moderately 100uA pullup is enough to drive the transistor, but I would have to crunch some numbers to know if will work in this circuit over the entire range (the low end) of beta values for a ‘3904.

Adding exteranl pullup is an easy solution in this case because this board’s power budget is not tight.

The tricky part is that quasi-directional ports have a read/write bit, giving the user the impression that it functions as either a push-pull output or a standard input.  The port’s ability to source some current furthers this misconception.

Getting More Power from Power-over-Ethernet (PoE)

Sunday, January 9th, 2011

A project involving high-power radios powered by PoE prompted me to look into the maximum power you can get from PoE.

IEEE 802.3af PoE – 15.4W max power
IEEE 802.3at PoE+ – 25.5W max power

Both types of PoE can be Mode A or Mode B.

Mode A – Power delivered on TX/RX pairs
Mode B – Power delivered on spare pairs

TI has a circuit to run Mode A and B simultaneously, resulting in a maximum power of 51W. The implementation is tricky because it must balance the current so that neither Mode A or B lines exceed the maximum current.

I would love to hear from someone who has implemented 51W PoE or used other cheats to get > 25.5W over Ethernt.

Low-Value High-Frequency Decoupling Caps Are a Myth

Tuesday, January 4th, 2011

In the past, I used three value-ranges of decoupling on high-speed designs:

  • Bulk: A 10uF or two caps located anywhere
  • Intermediate: Some 0.1uF caps located near the part
  • High-frequency: Some 1000pF located right under the BGA or connected to the power pins directly

You could not get high-value ceramics in small packages, so you had to use low-value capacitors to get the low inductance associated with a small package.

Now 0.1uF ceramics in an 0402 package are a standard low-cost part. I recently had a notion to check the curves to see at which frequencies a 1,000pF 0402 had a lower impedance than a 01.uF 0402:

There is only a tiny frequency range where the impedance of a 1,000pF is better. I would be interested to hear from anyone who knows a good reason to use them low-value decoupling caps today.