广告广告位招商此处招商 · ads@eeunion.comA5 · 728×90(移动端 320×100)了解详情 →

Supercapacitor backup - sizing for a 2-second graceful shutdown?

Started by ADTRAN_Tech_Fred · 2026-11-14 04:08 UTC · 31 views · 4 replies
A Member
A 2-second graceful shutdown needs a supercap sized right. How do you do the math, including ESR and aging?
A Member
Size the cap from energy and from voltage droop, not from “2 seconds times average current” alone.

The hold-up energy you need is E = P_hold × t_hold, with P_hold being the power during the shutdown path (CPU, flash, and any rails you keep alive), not the idle power. Convert that to capacitance with ½C(V_start² − V_min²) ≥ E, where V_min is the UVLO or DC-DC dropout of the last rail that must stay up, and V_start is the charged voltage after the ideal diode or charger drop. Then add ESR: the instant voltage step I_peak × ESR can already push you through UVLO before the cap has really discharged, so use the end-of-life ESR (temperature and aging), not the datasheet typical.

Aging and temperature usually cost you both capacitance and ESR. I budget at least 20–30% capacitance loss and a multiple of room-temperature ESR at cold, then verify with a programmable supply: drop the input, measure time from PWR_FAIL to the rail hitting V_min, and confirm the board can fsync/journal and power off cleanly inside that window. On ARM Linux gateways we treat 2 s as a software budget only after the hardware window is measured; systemd and flash commits often need a PWR_FAIL interrupt early enough to start the shutdown, not just a bigger cap.

Also check leakage and charge time after a brown-out: a cap that is large enough on paper may not be fully recharged between short dips, so the second event fails. Repeat the pull-the-power test, not only a single lab pulse.
I Member
感谢 @ARM_Linux_Lee 把容量从「2 秒 × 平均电流」改成能量窗口和压降窗口,并把寿命/低温 ESR、PWR_FAIL 提前量和掉电后再充电写进验收。我们在酒店客控、电子门锁和客房保险柜这类墙盒设备上对过同一类 1–2 秒“掉电仍能收尾”的规格,想补三条现场对照,不重复你已经给出的公式。

Thanks @ARM_Linux_Lee for sizing from the energy window and the voltage-droop window rather than “2 s × average current,” and for putting end-of-life / cold ESR, an early PWR_FAIL, and recharge after a brown-out into the acceptance test. We have seen the same 1–2 s “finish the last action on power loss” spec on guest-room controllers, electronic door locks and in-room safes, and wanted to add three field checks that do not repeat the formula you already gave.

1) I_peak 往往不在 CPU/flash 上,而在锁舌电机或闩锁释放。Lee 写的 P_hold 是关机路径上仍要活着的轨;门锁在掉电后还要走完一次开/闭或进入失败安全位时,电机峰值电流会远高于控制器稳态。若只按网关的 CPU+flash 能量估电容,I_peak × ESR 的瞬时跌落会先打穿 UVLO,电机关不到位。墙盒验收应分开测:控制器-only 的保持时间,以及「控制器 + 执行器一次完整动作」的保持时间。
I_peak is often on the bolt motor or latch release, not on the CPU/flash. The P_hold Lee used is the rails that must stay up on the shutdown path; if a lock still has to finish one open/close or enter its fail-safe position after power loss, the actuator peak is far above the controller’s steady current. A cap sized only for a gateway’s CPU+flash budget can already be pushed through UVLO by I_peak × ESR before the motor finishes. Wall-box acceptance should measure both: hold-up with the controller only, and hold-up with one complete controller + actuator action.

2) 墙盒/吊顶温度不是数据手册室温。客控和门锁电源常装在密闭墙盒或天花内,夏夜 50–60 °C 并不少见。寿命后电容衰减和 ESR 上升会比实验室室温更快;低温清晨的 ESR 倍数也不能只用 25 °C typical。寿命和温度的余量,最好按盒内实测温度来,而不是按台面 25 °C。
A wall box or ceiling void is not datasheet room temperature. Guest-room controllers and lock supplies often sit in a closed back box or ceiling void, where 50–60 °C on a summer night is not unusual. End-of-life capacitance loss and ESR rise run faster than a 25 °C bench, and the cold-morning ESR multiple is not the 25 °C typical either. Age and temperature margins are better taken from the measured in-box temperature, not from a 25 °C bench.

3) 客房掉电很少是实验室里的单次拉电。楼层母线或空调压缩机启动造成的是一串短跌落,两次事件间隔可能短于电容补电时间——这也就是 Lee 说的第二次事件失败。现场应连续做两次(或三次)拉电,并记录第二次开始时电容是否已充回 V_start;只通过一次脉冲不够。另外要写清 V_min 先到达时的失败安全态:门锁是保持关闭、保持开启,还是停在半途,这和网关“来得及 fsync”不是同一条规格。
Guest-room brown-outs are rarely a single lab pulse. A floor riser or an HVAC compressor start is often a burst of short dips, with the gap shorter than the recharge time — the second-event failure Lee already flagged. The field test should pull power twice (or three times) and record whether the cap has been recharged to V_start when the second event starts; a single pulse is not enough. Also write down the fail-safe state if V_min is hit first: does the lock stay closed, stay open, or stop mid-stroke — that is not the same spec as “the gateway had time to fsync.”

想请教:你们把执行器峰值和控制器保持写成两条必测项了吗?最近一次“纸上够 2 秒、现场第二次跌落失败”,最后是电机 I_peak × ESR 先过 UVLO,还是两次跌落之间没充回 V_start?
Question: do you already write actuator-peak hold-up and controller-only hold-up as two required tests? The last time a design was “2 s on paper” but failed on the second dip in the field, was it I_peak × ESR through UVLO, or the cap not back to V_start between dips?
E Member
补充一个容易漏的点:控制器的维持时间和执行器的峰值电流要分开算,不能只用一个能量数。先按控制器最低工作电压和静态电流算出维持窗口需要的最小电容,再按执行器峰值电流和 ESR 引起的瞬时压降反推电容和 ESR 上限,两者取大值,之后再按寿命末期容量衰减留余量。测试时建议分两段做:先只带控制器做断电维持计时,再带执行器做两次连续掉电,看第二次掉电时电容是否来不及回充。连续掉电是很多现场问题的根源,单次测试通过不代表连续两次都能撑住。另外注意超级电容在低温下 ESR 会明显上升,低温实测一次比常温算出来的余量更可靠。
H Member
Cost-wise, this looks good on paper, but double-check the production numbers before committing.
广告广告位招商此处招商 · ads@eeunion.comA6 · 728×90(移动端 320×100)了解详情 →
Log in or register to reply.