Path: ...!news.nobody.at!eternal-september.org!feeder3.eternal-september.org!news.eternal-september.org!.POSTED!not-for-mail
From: Don Y <blockedofcourse@foo.invalid>
Newsgroups: sci.electronics.design
Subject: Re: Ambient temperature control
Date: Mon, 1 Jul 2024 06:41:26 -0700
Organization: A noiseless patient Spider
Lines: 87
Message-ID: <v5ubm8$13cpk$1@dont-email.me>
References: <v5svtq$olhq$1@dont-email.me> <v5u3le$1209c$1@dont-email.me>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Injection-Date: Mon, 01 Jul 2024 15:41:29 +0200 (CEST)
Injection-Info: dont-email.me; posting-host="522e66ae658b05fa6b7048ee3045fe38";
	logging-data="1159988"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX1/Xgxlad1wxxu1L0eMUKd97"
User-Agent: Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:102.0) Gecko/20100101
 Thunderbird/102.2.2
Cancel-Lock: sha1:/L/wDVwUDbRUamqvFXcOrvz12yU=
Content-Language: en-US
In-Reply-To: <v5u3le$1209c$1@dont-email.me>
Bytes: 5355

On 7/1/2024 4:24 AM, Martin Brown wrote:
> On 01/07/2024 02:14, Don Y wrote:
>> Assuming you can keep a device in its "normal operating (temperature)
>> range", how advantageous is it (think MTBF) to drive that ambient
>> down?  And, is there a sweet spot (as there is a cost to lowering the
>> temperature)?
> 
> There can be for some high performance low level OPamps. Deliberately running 
> them as cold as is allowed helps take the LF noise floor down and by more than 
> you would predict from Johnson noise. ISTR there was a patent for doing this 
> back in the 1980's. Prior to that they tended to heat the front end to obtain 
> temperature stability and low drift.
> 
> https://ppubs.uspto.gov/dirsearch-public/print/downloadPdf/4883957
> 
> Made possible with the advent of decent solid state TECs.

We don't design products (industrial/consumer) that are "finicky" -- as
it leads to higher TCOs.  You don't want to need to control the environment
*or* have "skilled tradesmen" on staff to maintain/assure correct
performance.

The most common example (that I can think of) where temperature is
controlled FBO the electronics would be datacenters.  But, from the
research I've done, there, they simply set a desired temperature for
the cold aisle and largely ignore the resulting hot aisle temperature
(except to ensure it doesn't climb out-of-bounds).  I.e., they
don't close the loop on the hot aisle to control the cold aisle's
setpoint (cascaded control).

And, they don't get the cold aisle "as cold as possible" so they
acknowledge there are diminishing returns in doing so -- likely
cheaper to just pan on a (potentially) shorter upgrade cycle
than to waste electricity trying to eek out a bit more life.

Interestingly, I can't find anything other than "lore" to
explain why a *particular* cold aisle temperature is chosen.
Amusing to see how much folks DON'T know about the science
they apply!

When I designed my disk sanitizer, I did a fair bit of research
regarding temperature effects on drives -- because we process a
shitload (thousands) of *used* drives, annually and you don't want to
reuse a drive that has an increased chance of failure (based on
its previous environment, SMART data or observations while exercising
it).  The old "10 degree C" saw proved to be totally inappropriate,
*there*.

OTOH, I suspect it *is* worth noting for power supplies (as I
see most failures being toasted power supplies in otherwise
"healthy" products).  I suspect power *cycling* is a culprit, there
as I've seen failed solder joints where it looked like repeated
thermal expansion had led to the failure.

>> Also, is there any advantage to minimizing the hysteresis between
>> the ACTUAL operating temperature extremes in such a control strategy
>> (given that lower hysteresis usually comes at an increased cost)?
> 
> Depends how temperature sensitive the thing is that you are protecting. The 
> example I recall they were aiming for medium term stable 6 sig fig measurements 
> with the lowest possible noise.

I've needed to control temperature in applications where it was
key to the *process* being controlled.  E.g., monitoring exhaust
air temperature to determine the "state" of the bed and a cascade
loop on the inlet air handler to drive that to a desired state.

But, there, you have lots of money for the equipment and can buy
good/precise/fast control with things like face-and-bypass as
the primary controlled variable (so the control loop for the
heater/chiller can be cruder and more energy efficient).

"In the small", refrigeration is the only practical means of
lowering ambient temperatures.  And, that adds to operating costs.
If you can tolerate a wider deadband then the cooling cost
can be lower (e.g., cool to X degrees and let it *soak*, there,
before letting it warm to Y degrees instead of foolishly
trying to maintain the environment at some Z>X and <Y).
As you likely have LESS ability to precisely size the HVAC
to fit such a small load, deadband becomes a key consequence
of that selection process.

[Gotta wonder why data centers in northern latitudes don't
exploit outside air more agressively during the winter
months!]