Facebook
Google
GitHub
Linkedin
Hi everyone.
I spent about a million tokens on Claude Code for it to make me a spreadsheet that is complete and optimized and seeing the wonderful job it did - I can't even comprehend it all yet, I thought that a sharing wouldn't hurt, as I think it's too good to remain unknown.
So, the first thing, what does it's for?
It answers a design question: at what switching frequency does a DC-DC converter dissipate the least power, and - by extension - which components support that choice?
Input: You enter the topology (buck, boost, buck-boost, flyback, synchronous buck, or a manual mode for others), the operating point (Vin, Vout, Iout, L), and datasheet specifications for the MOSFET, driver, diode, inductor, and capacitors. Based on the topology, it automatically derives the duty cycle, blocking voltage, inductor current, and ripple-generating voltage - so you don't have to do the math by hand.
Model: It categorizes the eleven loss components into two opposing groups. Those that increase with frequency represent energy spent once per cycle: V×I overlap during transitions, Coss discharge, gate drive, diode reverse recovery, and body-diode conduction during dead time. Those that decrease with frequency stem from current ripple (defined as V_L·D/(L·f)): the lower the frequency, the higher the ripple and the greater the ohmic losses in the MOSFET channel, inductor copper, and capacitor ESR. Summing them yields a U-shaped curve - and the bottom of that curve represents the optimal frequency.
Output: The spreadsheet solves this in two ways, by design. The closed-form formula f* = (2·k₂/k₁)^(1/3) is derived from the calculus and reveals the problem's structure; the 61-point numerical sweep also accounts for core loss, which involves non-integer exponents and doesn't fit the formula. A significant divergence between the two provides valuable insight: it indicates that core or diode losses are dominating. The dashboard displays the optimal frequency, power loss, efficiency, and optimal inductance, along with seven alerts that invalidate the result if it falls outside the valid operating range - such as core saturation, entry into discontinuous mode, exceeding the thermal budget, or the minimum falling at the edge of the sweep range.
Beyond frequency: The same model is used to select the inductor (sweeping L at a fixed frequency), compare three candidate MOSFETs at the same operating point, measure the gain achieved by adopting ZVS, decide how many MOSFETs to place in parallel, and determine the frequency range that fits within a given thermal limit. A dedicated tab documents the physical basis of each equation, its assumptions, and the expected error - conduction calculations are reliable to within 10%, while switching calculations are engineering estimates with an error margin of 20% to 50%.
Why I did this: I really like the knowledge of current AI, it knows a lot, and also I always liked spreadsheets because of the versatility, usability, display format, graphs, and gradients you can do. I felt that it can really help people understand more, as I will try to learn reading the Derivation tab that explains how the equations were derived, and from what.
Hope you guys, and future people learning and trying to design such things use this, as it's free and I don't see any other place so complete (and for free).
Hope you enjoy, and give this post a like if you learnt something new.
Attached files: in English and in Portuguese (PT-BR).
Edit: forgot to say that I'm open to corrections, as I never worked with SMPS design, so my base is from college only. This spreadsheet isn't supposed to be an AI slop, if it is in some point, point it out and I'll search about it to address the changes and update it here.
I spent about a million tokens on Claude Code for it to make me a spreadsheet that is complete and optimized and seeing the wonderful job it did - I can't even comprehend it all yet, I thought that a sharing wouldn't hurt, as I think it's too good to remain unknown.
So, the first thing, what does it's for?
It answers a design question: at what switching frequency does a DC-DC converter dissipate the least power, and - by extension - which components support that choice?
Input: You enter the topology (buck, boost, buck-boost, flyback, synchronous buck, or a manual mode for others), the operating point (Vin, Vout, Iout, L), and datasheet specifications for the MOSFET, driver, diode, inductor, and capacitors. Based on the topology, it automatically derives the duty cycle, blocking voltage, inductor current, and ripple-generating voltage - so you don't have to do the math by hand.
Model: It categorizes the eleven loss components into two opposing groups. Those that increase with frequency represent energy spent once per cycle: V×I overlap during transitions, Coss discharge, gate drive, diode reverse recovery, and body-diode conduction during dead time. Those that decrease with frequency stem from current ripple (defined as V_L·D/(L·f)): the lower the frequency, the higher the ripple and the greater the ohmic losses in the MOSFET channel, inductor copper, and capacitor ESR. Summing them yields a U-shaped curve - and the bottom of that curve represents the optimal frequency.
Output: The spreadsheet solves this in two ways, by design. The closed-form formula f* = (2·k₂/k₁)^(1/3) is derived from the calculus and reveals the problem's structure; the 61-point numerical sweep also accounts for core loss, which involves non-integer exponents and doesn't fit the formula. A significant divergence between the two provides valuable insight: it indicates that core or diode losses are dominating. The dashboard displays the optimal frequency, power loss, efficiency, and optimal inductance, along with seven alerts that invalidate the result if it falls outside the valid operating range - such as core saturation, entry into discontinuous mode, exceeding the thermal budget, or the minimum falling at the edge of the sweep range.
Beyond frequency: The same model is used to select the inductor (sweeping L at a fixed frequency), compare three candidate MOSFETs at the same operating point, measure the gain achieved by adopting ZVS, decide how many MOSFETs to place in parallel, and determine the frequency range that fits within a given thermal limit. A dedicated tab documents the physical basis of each equation, its assumptions, and the expected error - conduction calculations are reliable to within 10%, while switching calculations are engineering estimates with an error margin of 20% to 50%.
Why I did this: I really like the knowledge of current AI, it knows a lot, and also I always liked spreadsheets because of the versatility, usability, display format, graphs, and gradients you can do. I felt that it can really help people understand more, as I will try to learn reading the Derivation tab that explains how the equations were derived, and from what.
Hope you guys, and future people learning and trying to design such things use this, as it's free and I don't see any other place so complete (and for free).
Hope you enjoy, and give this post a like if you learnt something new.
Attached files: in English and in Portuguese (PT-BR).
Edit: forgot to say that I'm open to corrections, as I never worked with SMPS design, so my base is from college only. This spreadsheet isn't supposed to be an AI slop, if it is in some point, point it out and I'll search about it to address the changes and update it here.
Last edited:
