------------------------------------------------ Part 1: Our boards operating conditions and failures Operating conditions: Vin:+5V+/-100mv, Vout is set to +3.3V. Input current is about 600mA so estimated Iout is about 800mA (at 90% efficiency) Failures description: First 2 failures occured at -30 celsius. Next one failure occured at normal temperature (20 to 30 celcius). We have an update! We've got one more failure but this time it's another board. Operating conditions are same, but Iout is about 1300mA. On new board all capacitors are of type X7R/X5R, Rfbt=26,7kOhm, Rcomp=20(Twenty)Ohm, Ccomp=2.2nF. Co=Total 780uF ceramic=100uF ceramic placed directly between IC output and ground pins (almost touches pins) plus 680uF spread over the board at the point of load. Summary Co ESR=0.76mOm Ci is ceramic 100uF but placed rather far from IC input (about 20mm). Failure occured at normal temperature (20 to 30 celcius). ------------------------------------------------ Part 2: Our boards compensation loop by SNVA417D rev.B Compensation loop for boards has been calculated long ago using App Note SNVA417D rev.B Reversed calculations gives following values: Rfbt Rcomp Ccomp Co_reverse ESR_reverse Co_nominal ESR_nominal Board#1 20k 0.33k 1000pF 333uF 1mOm 618uF 1mOm Board#1 26.7k 0.33k 2200pF 733uF 0.06mOm 780uF 0.76mOm ESR is calculated as product of all ESR on board, 6mOm per 100uF. Board#1 has 4x100uF, all other is a bunch of 4.7uF and 100nF capacitors. Board#2 has 7x100uF, all other is a bunch of 4.7uF and 100nF capacitors Vin is assumed as +5.5V in calculations to provide margins. Board#1 is calculated according to SNVA417D rev.B idealy. Capcitance derating should be in mind as it depends on temperature. For Board#1 it's 55% of nominal value (total derating for X7R/X5R is 65%). Well, there is sure mistake about ESR for Board#2, but as I can judge according to SNVA417D it shouldn't be the case of failure because with this compensation values knee of 40dB/dec to 20dB/dec is virtualy moved to higher frequency (3.6MHz to be exact) Also there is not much capacitance derating for Board#2. Please note that if do reverse calculations using compensation values, recomended by SNVA417D rev.B itself, Co_reverse is about 55%-60% of Co_nominal, exactly as in our case for Board#1. Also when I calculated ESR for recomended values it was at high boundary of nominal range (note this, I would refer to it in Part 4). ------------------------------------------------ Part 3: Our boards compensation loop as it 'seen' by SNVA417D rev.D Let's see if we use same compensation values and do reverse calculations by equations of App Note SNVA417D rev.D Rfbt Rcomp Ccomp Co_reverse ESR_reverse Co_nominal ESR_nominal Board#1 20k 0.33k 1000pF 187uF 1.76mOm 618uF 1mOm Board#2 26.7k 0.33k 2200pF 1570uF 0.03mOm 780uF 0.76mOm As you can see, after changing calculation method: Co_reverse for Board#1 went low about 2 times, and ESR went high about 2 times Co_reverse for Board#2 went HIGH about 2 times and ESR went LOW 2 times Changes for Board#2 are opposite to Board#1!!! It bothers me much!!! In both cases Co_reverse have much difference from Co_nominal. For Board#1 it's 3 times less than nominal, for Board#2 it's 2 times larger than nominal. Co_reverse is calculated like fLC=fCOMP fCOMP=1/(2*pi*Ccomp*(Rfbt+Rcomp)) Co=1/( (2*pi*fLC)^2 * L )=1/( (2*pi*fCOMP)^2 * L ) ESR_reverse is calculated like fESR=fPOLE fPOLE=1/(2*pi*Ccomp*Rcomp) Resr=1/(2*pi*fESR*Co)=1/(2*pi*fPOLE*Co) Proceeding further with reverse calculations i found following: As it's stated in SNVA417D rev.D cross-over frequency (fX) is 1/10 of fSW=1MHz/10=100kHz. Let's assume that Iout=3A, so with Vout=+3.3V Rout would be 1.1Ohm. Vin is +5.5V to have a margin. In the end i reversly calculated Rcint and it should have value of 4.03kOhm for Board#1, and 4.86kOhm for Board#2 which is wrong as you have said that Rcint is 105.8kOhm so this probably must be root of difference between rev.B and rev.D of SNVA417D app note. ------------------------------------------------ Part 4: Reverse calculations for recomended values in SNVA417D rev.D I spent my time and done revers calculations for recomended values of App Note SNVA417D rev.D for LMZ10503 and Vin=5V, here are results: Rfbt Rcomp Ccomp Co_reverse ESR_reverse Rcint_reverse Co_nominal ESR_nominal Co_derating Row#1 150k 1k 47pF 22.89uF 2.05mOm 24,41k 22uF 2-20 mOm 0% Row#2 100k 4.53k 100pF 49.67uF 9.12mOm 20,24k 47uF 2-20 mOm 0% Row#3 71.5k 2k 180pF 79.56uF 4.52mOm 14,20k 100uF 1-10 mOm 20% Row#4 56.2k 0.499k 270pF 106.53uF 1.26mOm 10,15k 150uF 1-5 mOm 30% Row#5 100k 4.53k 180pF 160.92uF 5.07mOm 23,01k 150uF 10-25 mOm -7% !!! Row#6 182k 8.25k 100pF 164.52uF 5.01mOm 42,07k 150uF 26-50 mOm -10% !!! Row#7 133k 4.99k 160pF 221.57uF 3.60mOm 30,62k 220uF 15-30 mOm 0% Row#8 200k 6.98k 100pF 194.73uF 3.58mOm 44,99k 220uF 31-60 mOm 10% What is interesting: 1: No margin for derating in Co_reverse. Only Row#3, Row#4 and Row#8 have some margin for capcitance drop but not that much as rev B. Rows #5 and #6 even violates nominal Co 2: Rcint is floating about 20kOhm +/-10kOhm. Only for Row#6 and Row#8 Rcint is about 40kOhm which still isn't 105.8kOhm 3: ESR_reverse is at low boundary - to center of nominal range for rows #1-#4. For #5-#8 (large capacitance) it falls out of range and is less than nominal. When I reversed rev.B (using rev.B equations) ESR_reverse was at high boundary of nominal range for all rows (even a bit higher). ------------------------------------------------ Part 5a: New compensation loop values for our boards Using SNVA417D rev.D I have calculated new loop compensation values: Co_nominal ESR_nominal Rfbt Rcomp Ccomp Board#1 618uF 1mOm 467k 7.97k 78pF Board#1 780uF 0.76mOm 459k 6.67k 90pF Well... absolutely different thing if compare to initial ones. I reverse calculated loop compensation values with rev.D equations and C_reverse and ESR_reverse are much close to Co_nominal and ESR_nominal, so I can judge that calculations are sane. But if I use this values and do reverse calculated with rev.B equations: Rfbt Rcomp Ccomp Co_reverse ESR_reverse Co_nominal ESR_nominal Board#1 467k 7.97k 78pF 26uF 24mOm 618uF 1mOm Board#1 459k 6.67k 90pF 30uF 20mOm 780uF 0.76mOm and it looks like catastrophe! So using new values for compensation loop scares me much! I'll try them with WEBENCH soon and give report... ------------------------------------------------ Part 5b: New compensation loop values for our boards with Rcint=17k As I denoted above - Rcint with recomended values was about 20k so I tried to use this value for calculations. Following data is same as 5a but calculated with Rcint=17k (I choose that value as with it Rfbt would be 75k, which is common value for Rfbt in LMZ10503 datasheet for Rfbb recomendations) Co_nominal ESR_nominal Rfbt Rcomp Ccomp Board#1 618uF 1mOm 75k 1.28k 482pF Board#1 780uF 0.76mOm 73.8k 1.07k 553pF And if I use this values and do reverse calculated with rev.B equations: Rfbt Rcomp Ccomp Co_reverse ESR_reverse Co_nominal ESR_nominal Board#1 75k 1.28k 482pF 160uF 3.84mOm 618uF 1mOm Board#1 73.8k 1.07k 553pF 184uF 3.21mOm 780uF 0.76mOm ------------------------------------------------ Part 6: Sanity check Also I tried to reverse recomended values of rev.D using rev.B equations and I've got: Rfbt Rcomp Ccomp Co_reverse ESR_reverse Co_nominal ESR_nominal Row#1 150k 1k 47pF 15.6uF 3.0mOm 22uF 2-20 mOm Row#2 100k 4.53k 100pF 33.33uF 13.59mOm 47uF 2-20 mOm Row#3 71.5k 2k 180pF 60.00uF 6.0mOm 100uF 1-10 mOm Row#4 56.2k 0.499k 270pF 90.00uF 1.5mOm 150uF 1-5 mOm Row#5 100k 4.53k 180pF 60.00uF 13.59mOm 150uF 10-25 mOm Row#6 182k 8.25k 100pF 33.33uF 24.75mOm 150uF 26-50 mOm Row#7 133k 4.99k 160pF 53.33uF 14.97mOm 220uF 15-30 mOm Row#8 200k 6.98k 100pF 33.33uF 20.94mOm 220uF 31-60 mOm Well - good hits for ESR for all rows and good hits for Co for rows #1-#4 (capacitance derating margin is at 30-40% like with initial rev.B recomendations) BUT realy realy bad for large capacitance (rows #5-#8) - reversly calculated capacitance is 15% to 25% of nominal value Also note, that with high capacitance proportion of Co_reversed/Co_nominal values looks like the one I got in Part 5.b when used 17k for Rcint. ------------------------------------------------ Part 7: Questions about calculations =========== QUESTION 1. If I'm using Rev.D for calculations should I derate capacitance? If nominal Co is 618uF and it have total precision/fluctuation from -32% to +38% over operating temperature range then what value should be used? (I have not seen derating for most recomended values in rev.D, look at Part 4) =========== QUESTION 2. What Rcint should be used in calculations? 105.8kOhm or 17kOhm? I already tried values of Part 5b for Board#1 in WEBENCH and I've got realy good results. =========== QUESTION 3. Was rev.B recomendations that insane for high output capacitance? Could this be root of our problems? =========== ------------------------------------------------ Part 8a: Experiments - make it unstable We did some experiments to make LMZ10503 working unstable, but no success What we tried: Most experiments were done on wireboard that imitates real board - Cin=100uF, Vin=+5V, Vout=+3.3V, Iout changes form uA to 1.3A (uA,330mA,660mA,1.3A), Cout changes from 100uF to 700uF (100uF, 300uF, 500uF, 700uF). 1. Reduce Co to 100uF on a wireboard using existing compensation loop. 2. Changed Ccomp on a wireboard from 1nF downto 10pF (Rcomp=330Ohm). 3. Changed Rcomp on a wireboard from 330Ohm up to 510kOhm (Ccomp=1nF). On Real board we done capacitance derating imitation. 4. Reduce Co to 265uF on a REAL board using existing compensation loop (removed all 100uF and placed one 47uF at IC output pin, 4.7uF and 100nf still scattered over board). 5. On real board we had increased path from Vout/Ground to nearest Cout(47 uF) and made it 50mm long. Most worse effect we've got was sinusoidal voltage regulation with amplitude of 90mV (3% of Vout) in experiment 2. Nothing could make it working unstable - changing Iout on the fly from 0 to 1,3A and 330mA to 1,3A in a single step (other combinatinos also went fine). Part recovered easily after connecting to the output discharged 600uF capacitance on the fly (there were voltage drop and after some time voltage rised back slowly without excessive overvoltage and generation events). ------------------------------------------------ Part 8b: Experiments - short circuit to the ground Well, that thing burned IC at last. On the fly Vout network was shortened to the ground and after some time (about 5 seconds) short circuit condition was removed. Being shortened IC went into current limiting mode (if judge by the input current - it was 0.53A or 0.85A, it's pitty but probably there was mistake in observations...) After short circuit condition were removed IC has recovered and rised voltage on Vout to it's nominal value (+3.3V). After about 10 repetitions we observed high input current. After turning Vin off we had measured and found short circuit within IC between Vout pin and Ground (almost Zero Ohm). At this time there were no short circuit between Vin and Ground. We had applied input voltage back. After some time Vin pin also became shortened to the ground. =========== QUESTION 4: Why nor short circuit protection nor overtemperature protection haven't saved IC? ===========