How the professionals do it?

Given a certain spec (like for the LM 337):

Min I load : typical 2,5 mA - max 10 mA

Should I calculate my circuit for those 10 mA? Or hope it is 2,5 mA?

Please note that my question is not about the regulator but about how a correct design should be done.

 

From Onsemi: "This imposes the requirement for a minimum load current. If the load current is less than this minimum, the output voltage will rise."

Since the maximum value for the minimum load is specified for 10mA I would use that value. If you use 2.5mA and it is actually 10mA, the voltage could rise (as described above).

 

Typ is what one might expect from a random sampling of devices.

One would design for worst case scenario and this could be either min or max specification
depending on specific parameter in question.

 

If you design using "typical" specs then some of your projects might work.
But if you design using "worst case" specs then all of your projects will work.

I don't believe "typical" specs. What if your supplier has run out of a part then ordered a few hundred more? The manufacturer sends the few hundred from a single production run that is not a random sample. They all might be low, medium or high.

What if somebody ordered "the best ones" from a manufacturer and only the worst ones remain for you to buy?

 

Thanks to you all for replying. Quite clear!

Additional conclusion is that typical values seem not relevant provided that I have to catter for worst case.

 

I worked as an applications engineer in several semiconductor companies, and typical specs have no meaning. They are often conjured up by marketing to gain a competitive edge and have no relation to product performance. Quality companies have rules that put some controls on typical specs, but designing with typical specs is like playing Russian roulette with 6 bullets in the gun, bad news is guaranteed.

 

As a working design engineer, my answer is ...

It depends on the parameter and how it affects the rest of your design. If its a critical parameter that could cause a part of the design to fail or perform poorly, then use the worst case specs. In other cases where its not as important, use typical.

If you want to be a good designer you have to understand ALL of the implications of your design. Use typical or worst case as it makes sense in your circuit. Calulate everything you can - test everything!

 

Calulate everything you can - test everything!

I built all my prototypes on stripboard (Veroboard). Then there were no intermittent breadboard contacts nor too much stray capacitance.

I designed the circuit then layed it out on the stripboard so that the strips were all cut to length then re-used many times. The copper strips formed half or more of a pcb and the parts were the remainder. A few jumper wires were used.
Most of my stripboard prototypes were sold.

I designed a product and had thousands made. Almost EVERY one worked perfectly.
I DID NOT test all the parts, instead I trusted the Name Brand local manufacturers and used their worst case spec's on their datasheets, not ebay counterfeight junk.

Only two out of thousands failed before they were sold. One had its IC installed backwards and the other had a shorted electrolytic capacitor.
None failed after sale.

 

Let me clarify ... I didn't mean to test everything in production. I meant to do thorough design verification testing on the design to make sure that your design is working as expected.

If you have to 100% test everything in production, then you have a marginal circuit, or some bad part suppliers.

 

What? You don't test 100% of everything before you ship?

 

I made only about 30,000 of one product. Every one was thoroughly tested before shipment.
Two failed.
If I made one million or more then maybe I would test only 1 out of 5.

 

Test each item ... but not 100% of it specifications

 

Test each item ... but not 100% of it specifications

I designed and had made more than 30,000 items of a PRO audio equalizer product.
Its spec's were plus and minus 3dB, distortion less than 0.05% and noise less than -60dB. ALL were tested.

EDIT:
Good quality control tests EVERYTHING. Poor quality control has nothing to do with quality.

 

Thats great Audioguru. Congratulations on a successful product.

Some products lend themselves to 100% testing and some don't. Before I was a designer, I was a test engineer. Some products are so complex and have so many features, that 100% testing of every feature, function and specification is not practical. You have to analyze the design and determine what needs to be tested and what doesn't. This is industry standard. You have to put out a working product, and you have to do it in a resonable time, and you have to make money.

 

Thats great Audioguru. Congratulations on a successful product.

Thanks.
It was for a Japanese PRO audio manufacturer. Mine was designed and made by me in Canada. When they copied my design and made it in Japan one year later then many failed, hee, hee.

 

Today it would be China, but the same thing goes on. Circuits and even entire products are copied. Problem is that the people copying the design often don't really know how it works and have troube duplicating it.

 

Run it mid way between max and typical.

 

Run it mid way between max and typical.

Then the thousands or millions of parts that are minimum WILL NOT WORK!

 

Run it mid way between max and typical.

I agree with AG. This makes no sense at all.

 

I designed hundreds of projects in 40 years and here is the benefit of my experience. The amount of conservatism used in the design is based on the projected design volume. Discrete gates for computers (1960s) were very high volume and the worst case specs were our min/max specs at our operating points; not manufacturer's specs. Medium volume designs used worst case min/max specs and allowed for temp changes, life, etc. Low volume designs used min/max specs. Typical specs were never used because they varied so much that you had to lock into one vendor for them to be valid. When min/max specs were not available they were generated using statistics and purchased samples. Imagine, A 5% resistor has a 10% tolerance over life!

As a semiconductor applications engineer I often dealt with circuits that were designed long ago and now malfunctioned (or copied). Sorry, but the problems almost always resulted from the use of typical specs, depending on non-specified parameters, or misapplication of the part. Being cute and tricky only works for a while. When semiconductor companies change lines or processes on a part they test for min/max parameters not typicals. Only a newbie or an extreme optimist takes typical specs serious on a 30 year old part that has been changed many times .

Using min/max specs for design and allowing for life and ambient changes seldom costs much if a design is done by a skilled engineer.

I have always tested every product, but not every function. Skilled test engineers can relate functions so that 100% testing of each product is not required.