Hardware testing is a difficult problem

Nubility

Hardware testing is a difficult problem [Copy link]

 

I used to think that hardware testing was a very simple job, but recently I encountered several very difficult testing problems and found that it is not easy to achieve safe and reliable testing.

1. For high current measurement, the customer requires that a current sensing resistor must be placed to indirectly measure the current (I think this is stupid), but in fact, the current of tens of amperes is drawn out from the inner layer through the vias and then returned to the inner layer through the current sensing resistor. A large part of the voltage falls on the current sensing resistor and the vias, and the expected current cannot be reached at all. I feel that this problem must be solved by using a DCDC chip with built-in current detection

2. For the power-on timing test, there are 17 power supplies on the board that need to determine the timing relationship. I only have a dual-channel oscilloscope so I test two at a time, but I was criticized. Some people said that this is not accurate and that it is reliable to test all the power supplies at the same time. But where can I find a 17-channel oscilloscope? That being said, if I really want to test the timing of 17 power supplies at the same time, the only thing I can think of is to connect these power supplies to the FPGA to compare the time difference between them.

3. Ripple test of multiple power supplies. Some CPUs have many types of power supplies and the pins are very dense. Many filter capacitors on the back of the PCB are 0402 or 0201 in size. When manually measuring with an oscilloscope probe, the ground spring probe is very easy to deviate and cause a short circuit. Not long ago, several DCDCs were ruined because of the spring needle. If it is a more sophisticated CPU, the filter capacitor uses a 01005 package. How should it be tested? Can a test fixture be made and pressed directly on the PCB to safely contact all the signals to be tested?

laker2008

Um... LZ's product is so complicated, it actually has 17 power supplies, which is beyond my imagination

Nubility

laker2008 posted on 2023-8-22 17:30 Um... LZ's product is so complicated, it actually has 17 power supplies, which is beyond my imagination

Many of them have the same voltage, but they have different uses. You can count them on one hand if you classify them by voltage.

wyhlhywyb2021

Testing requires extensive testing experience and basic knowledge.

zhaoyanhao

What CPU is used? I don't think all 17 power supplies need to be tested for timing. Is it that Party A or someone else needs it for the sake of demand? There is no need to test the timing of 17 power supplies.

Nubility

zhaoyanhao posted on 2023-8-22 17:44 What CPU do you use? I don't think all 17 power supplies need to be tested for timing. Is it because Party A or someone else needs it for the sake of demand? There is no need to test 17...

I don’t know. This CPU was successfully taped out for the first time. The customer was very picky and knew the answer without asking.

freebsder

Finding a problem is always more troublesome than starting a new one.

秦天qintian0303

Now it has become automated testing, various automatic inspection

大秦正声

If the design is not good, there is no way to test it. If the device package is small, welding will be a problem! Debugging and mass production are all pitfalls!

qwqwqw2088

Hardware testing is sometimes related to test instruments

Can I use a logic analyzer to measure the timing of 17 power supplies? But is it necessary to measure the power supply in this way? It is not an FPGA signal.

Nubility

qwqwqw2088 posted on 2023-8-23 15:36 Hardware testing is sometimes related to test instruments. Can the timing of 17 power supplies be measured with a logic analyzer? But is it necessary to test the power supply in this way? ...

Power-on sequence requirements

xutong

If there are many paths, you must use a logic analyzer. It is a bit outrageous to use an oscilloscope to measure them one by one.

It is okay to measure the current of tens of A with a shunt resistor, even if it is 1mR30A, it will only be 30mV

Nubility

xutong posted on 2023-8-25 17:26 If there are many paths, you must use a logic analyzer. It is a bit outrageous to measure them one by one with an oscilloscope. It is okay to measure the current of dozens of A with a shunt resistor. Even if it is 1mR30A, it is only 30mV...

I have never used a logic analyzer. Does it support wide voltage?

xutong

Nubility posted on 2023-8-25 17:43 I have never used a logic analyzer. Does it support wide voltage?

The resistor voltage is 5V.

Nubility

xutong posted on 2023-8-25 17:56 It can be divided by resistors, usually 5V

The power supplies here are all 0.8v, and most of them are 1.2v.

xutong

Nubility published on 2023-8-25 18:45 The power supplies here are all 0.8v, and most are 1.2v

There's no need for pressure division.

wsmysyn

Big chips are complicated, with all kinds of power supplies, which is normal. If you have ATE equipment, everything will be easy. The disadvantage is that it is expensive.

I am currently working on a BGA3700 multi-ball chip mass production test solution.

The TDP power consumption of a single die is more than 300W.

Nubility

wsmysyn posted on 2023-8-25 19:59 Big chips are complicated, with all kinds of power supplies. This is normal. If you have ATE equipment, everything is easy. The disadvantage is that it is expensive. I am now working on a BGA3700...

Server CPU?

Nubility

wsmysyn posted on 2023-8-25 19: 59Big chips are complicated, with various power supplies, which is normal. If you have ATE equipment, everything is easy. The disadvantage is that it is expensive. I am currently working on a BGA3700...

No, even a 100M oscilloscope cannot be used by two people.

wsmysyn

Nubility posted on 2023-8-25 20:53 Server CPU?

Server computing chips