Q: Why comparator datasheet no max/min value rise time

Thread Starter

JBmtk

Joined Jun 29, 2009
23
I am in this testing class and one of the questions was why there are no max/min limits set for the rise/fall as well as the rest of the timing datasheet specs.

The chip a LT6703 comparator, but after looking at multiple other comparators, it seems that there is a consistent pattern

There is a typical value, but I don't understand that even though there are testing conditions giving, there is no max or min value to test for.

I was thinking it could maybe due to hysteresis, but there is a test limit for that.
 

Papabravo

Joined Feb 24, 2006
21,157
The answer has to do with the methodology of making the determination. You might be tempted to think that min and max values are absolute boundaries. This is probably not the case. Min and max values are determined by statistical sampling and represent the ±3σ points about the mean of a normal distribution.

Making these tests for for each parameter is time consuming and expensive. It is my view that in most cases, if your design depends on specific values of one or more parameters then your design has a fatal weakness. A manufacturer may decide not to invest the time and resources in gathering and presenting such data for each and every measurable parameter.

In RF work we seldom depend on manufacturer data. We spend great amounts of time characterizing new devices so we understand what we might expect from a production lot. In our designs it is absolutely essential to minimize the sensitivity to parameter variations. As good as it is semiconductor manufacturing is far from perfect.

According to Murphy's law "All parameters are random variables with a hopefully small variance"
 

kdillinger

Joined Jul 26, 2009
141
The answer has to do with the methodology of making the determination. You might be tempted to think that min and max values are absolute boundaries. This is probably not the case. Min and max values are determined by statistical sampling and represent the ±3σ points about the mean of a normal distribution.
For the purpose of yield, they are absolute boundaries. When silicon is tested according to the final production program any device that fails the MIN/MAX limits is rejected. This ensures that all devices that come out of production will be guaranteed to meet the MIN/MAX spec.

Making these tests for for each parameter is time consuming and expensive. It is my view that in most cases, if your design depends on specific values of one or more parameters then your design has a fatal weakness. A manufacturer may decide not to invest the time and resources in gathering and presenting such data for each and every measurable parameter.
People would disagree with your "fatal weakness" statement. Myself, and other would agree, that your best friend is the MIN/MAX spec in the datasheet because as I previously stated, those are guaranteed by the manufacturer. Typicals are what one should be careful of. When it so happens a critical paramater in ones design hinges on a parameter that is only listed as 'typical' these can lead to failures and may be a poor design (ask yourself are you making toy planes or mission critical sensors on an automobile).

In RF work we seldom depend on manufacturer data. We spend great amounts of time characterizing new devices so we understand what we might expect from a production lot. In our designs it is absolutely essential to minimize the sensitivity to parameter variations. As good as it is semiconductor manufacturing is far from perfect.
Then you do extra screening beyond the datasheet and this is another design philosophy. Some people design to 'worst case' meaning they heed the MIN/MAX and ensure that their design can handle shifts to these guaranteed extremes.
In my experience, extra screening is for designers who desire tight control over given parameters. Just as expensive and time consuming as production testing.
 

Papabravo

Joined Feb 24, 2006
21,157
Your first response indicates your belief that every chip that comes off the line is tested for compliance. I do not believe that is the case -- do you?

So what is your explanation for a data sheet with a typical value and no MIN/MAX specification? How would you do a "worst case" analysis with such a part?
 

retched

Joined Dec 5, 2009
5,207
I am in this testing class and one of the questions was why there are no max/min limits set for the rise/fall as well as the rest of the timing datasheet specs.
kdillinger, I understand your thoughts, but they do not pertain to this. This thread is about the LACK of max/min in a datasheet, not about heeding the max/min.
 

kdillinger

Joined Jul 26, 2009
141
Your first response indicates your belief that every chip that comes off the line is tested for compliance. I do not believe that is the case -- do you?
Absolutely. Every device has a final production test. This is how semiconductor manufacturers keep track of yield percentages from a wafer lot.

This does not mean that every parameter is tested, but every device is tested nonetheless.

So what is your explanation for a data sheet with a typical value and no MIN/MAX specification? How would you do a "worst case" analysis with such a part?
1) Typical values are usually covered during characterization which is separate from final production testing. Typical parameters may certainly be tested during final test to monitor yield for example, or they may simply be tested during characterization.
If a parameter only has a typical value it may have not been critical for the customer's application. Semiconductors are born from requirements of the customer.
In terms of worst case design for a typical; easy you can't. I never said that 'worst case' design is an absolute. This is where one finds a manufacturer willing to do additional testing themselves, willing to explain the innards of the device to give a warm fuzzy (over temp, etc.), find a device that does list the critical parameter as a MIN/MAX, or do screening oneself.
 

kdillinger

Joined Jul 26, 2009
141
Your reply is most easily summed up as "I don't know" which is ok. I been around the industry enough and know people in the industry, an expert would be a Product Engineer, that every device has a final test program. It is how semiconductor manufactures keep track of process shifts and yield.

More to the point to the OP's question, I can give an example. THD+N vs. Frequency is a test that takes too much time in a production test environment, say 1 second. This is an eternity. Typical TOTAL test times are roughly in the milliseconds and even then that is extremely long. For certain devices, those that are manufactured at millions per month, test time counts. Long test times mean more money spent.

This is but one reason why certain parameters may not have a MIN/MAX spec.

The real answer is you have to talk to the particular PE in a particular group in a given company to know why one parameter has a MIN/MAX while others only have a TYP.

Not an opinion. Fact based on experience.
 

retched

Joined Dec 5, 2009
5,207
Now, I undestand that EVERY device cant go through EVERY test during production. But, is ONE taken and tested for 1 second, and the results published somewhere?

If the op wanted to see, on paper the min/max of a device, that wasn't in the datasheet, where would he look? Would there be an app note that would say, "Because Device X was found to have a max/min of w, this works: and this doesn't:
?
 

kdillinger

Joined Jul 26, 2009
141
and the results published somewhere?
Yes. Internally to the company.

BTW, there are usually 3 types of testing:

ATE Characterization.
Bench Characterization.
Final Production Test - which have the limits set by the data pulled from the ATE and/or Bench characterization.

If the op wanted to see, on paper the min/max of a device, that wasn't in the datasheet, where would he look?
Ask the manufacturer if they have that data. It there is no MIN/MAX then it may have not been tested because of long test times which most likely means it is not a critical parameter. Being a critical parameter is the overriding factor.
 
Last edited:

Papabravo

Joined Feb 24, 2006
21,157
Your reply is most easily summed up as "I don't know" which is ok. I been around the industry enough and know people in the industry, an expert would be a Product Engineer, that every device has a final test program. It is how semiconductor manufactures keep track of process shifts and yield.

More to the point to the OP's question, I can give an example. THD+N vs. Frequency is a test that takes too much time in a production test environment, say 1 second. This is an eternity. Typical TOTAL test times are roughly in the milliseconds and even then that is extremely long. For certain devices, those that are manufactured at millions per month, test time counts. Long test times mean more money spent.

This is but one reason why certain parameters may not have a MIN/MAX spec.

The real answer is you have to talk to the particular PE in a particular group in a given company to know why one parameter has a MIN/MAX while others only have a TYP.

Not an opinion. Fact based on experience.
What I do know is that no company could stay in business if they tested all the parameters of every unit that comes off the line. I think you ended up saying the same thing so we are in agreement.
 
Top