Fixed Frequency Control vs Constant OnTime Control of

Fixed Frequency Control vs Constant On-Time Control of Step-Down Converters Voltage-mode/Current-mode vs D-CAP 2™/D-CAP 3™ Masashi Nogawa Sr. Systems Engineer, SWIFT team, Power Management 1

Contents • Abbreviation/Acronym • Quick history • How each control mode works • Large signal transient response • Line transient response • Loop stability (VM / CM / HM) – Bode plot – Output impedance plot – Small signal transient response • Output voltage noise 2

BUCK CONVERTER BASIC 20160426 3

Common Operation of Step-Down (Buck) Converters VM/CM: PWM Generator HM: ON pulse Generator Equation at Stable Condition D VIN = VOUT With HM, Take “D” as pulse “Density” 20160426 Simulation Waveform 4

Buck Conv. Common Operation: Switching Frequency • Simulation Waveform 20160426 5

Duty/Density Change VM/CM vs HM • Force changing “D” control signal to monitor VOUT response. • VOUT will just follow “D” over constant VIN. D VIN = VOUT f. SW goes down approaching to f. LC Simulation Waveform 20160426 6

BUCK CONVERTER COMPENSATION 20160426 7

Buck Conv. Common without any Loop Compensation • All buck-converter without any loop compensation is guaranteed to oscillate Need to compensate properly for VM, CM, and HM. VOUT goes into negative input of controller 20160426 Start the loop Pulse sequence converted to IL IL goes into output capacitor 8

Loop Compensation: Voltage Mode Figure 7, TI Apps Note SLVA 662 • Type-III compensation is popular, but tough to design • The compensator re-shape feedback signal to boost phase back. Completely free from handling output current information (unlike CM or HM). Always the same response over the same VIN and VOUT condition. 20160426 9

Loop Compensation: Current Mode • Double-loop structure: the inner current-loop masks inductor from the main voltage-loop. Easy to calculate compensation component The loop is a heavy function of output current. Double-loop limits freq. configuration: f. SW > f. CURRENT_LOOP > f. VOLTAGE_LOOP With x 10 rule, the main loop will be x 100 slower than f. SW 20160426 TI Apps Note SLVA 352 A 10

Loop Compensation: Hysteretic Mode • Ripple injection method is popular to compensate. • Basic idea is override feedback signal (could oscillate) with an output current “ripple” information. This is equivalent to generate a virtual ESR, only at feedback path. Heavy ripple injection leads to the system similar to current-mode. ideal case Many devices are internally compensated. 20160426 11

Almighty Loop Compensation (Debugging Only) • Disregarding all other factors, old-days electrolytic capacitors or Tantalum capacitors will bring ultimate stability due to their high ESR. • Frequency region above RESR and COUT resonant frequency, the output capacitor turns into resistor, so no 90 shift. High ESR system generates huge amount of ripple that is not acceptable in most of applications. …not practical… This method is very good for a debugging of unknown oscillation. 20160426 12

APPLE-TO-APPLE COMPARISON 20160426 13

Comparison Condition / Representative Parts Parameter Input Voltage Range (VIN) Output Voltage (VOUT) Maximum Output Current (IOUT) Temperature Switching Frequency (f. SW) 20160426 Value Note 5 V to 12 V Select a device to work at both 5 V and 12 V 1. 2 V Popular voltage rail for many logic ICs 5 A Room Temp Just an ease of comparison, don’t take high- or low- temperature data in this document Range of Analog ICs can’t hit completely 600 k. Hz to the same frequency. Set 700 k. Hz samples in narrow enough range. Kept the original EVM design as much as possible Mode Referred as Control mode detail Compensation Input Voltage Range Output Current Max High-side FET Low-side FET Frequency Inductor Output Capacitor L x Cout Resonant Freq. Input Capacitor Voltage-mode VM-IC Voltage-Mode with Input Voltage Feed-forward Type-III Current-mode CM-IC 2 Peak-Current-Mode Hysteretic-mode HM-IC D-CAP 2 Type-II 4. 5 V to 16 V 4. 5 V to 17 V Internal Ripple Injection 4. 5 V to 18 V 6 A 6 A 6 A 26 m (Integrated) 13 m (Integrated) 600 k. Hz 1 u. H 22 u. F 5 (Ceramic) 15. 2 k. Hz 26 m (Integrated) 36 m (Integrated) 19 m (Integrated) 28 m (Integrated) 633 k. Hz 1. 2 MHz 688 k. Hz 3. 3 u. H 1. 5 u. H 200 u. F 220 u. F (Tantalum) 22 u. F 2 (Ceramic) 6. 20 k. Hz 5. 91 k. Hz 19. 6 k. Hz To eliminate input impedance impact on the measurements, wide (frequency) range of bypass capacitors are connected at each board edge. *1: for further comparisons, 2 nd configuration of CM-IC is prepared whose EVM has a higher switching frequency setting. 14

Apple-to-apple comparison LARGE SIGNAL LOAD TRANSIENT RESPONSE PART 1: BASIC COMPARISON 20160426 15

Load Transient VM/CM vs HM Consuming Charge on COUT Re-Charge COUT VM / CM Constant Frequency Pulse 20160426 Simulation Waveform Consuming Charge on COUT Re-Charge COUT HM Variable Frequency Pulse Constant TON 16

Large Signal Load Transient Response 30 ns/A slew rate 500 m. A 5 A 500 m. A See next slide 20160426 17

Large Signal Load Transient Response: Undershoot VIN Feed-forward, VM-IC: D VIN = VOUT No VIN difference, no ringing. Recovery time ~ 15 us 2 loops: VERR IL VOUT CM-IC: Inner current-loop ringing before VOUT recovery. Recovery time ~ 15 us HM-IC: Only Comparator VOUT > VTARGET Ringing and bigger undershoot (less COUT) Recovery time ~ 5 us-10 us 20160426 18

Apple-to-apple comparison LARGE SIGNAL LOAD TRANSIENT RESPONSE PART 2: COMPARISONS VARIOUS PARAMETERS 20160426 19

Large Signal Load Transient Response: IOUT Dependency VM-IC: No IOUT difference, only amplitude changes. Recovery time ~ 15 us All control show the same ringing behaviors, CM-IC: just a different amplitude. Inner current-loop ringing before VOUT (CM has different DC gain but can’t be recovery. observed here. ) Recovery time ~ 15 us HM-IC: DC voltage error (light-load mode) Recovery time ~ 4 us 20160426 20

Large Signal Load Transient Response: COUT = 1 m. F VM-IC: Recovery time ~ 60 us CM-IC: Recovery time ~ 50 us HM-IC: Recovery time ~ 30 us 20160426 21

Large Signal Load Transient Response: Pulse Behavior Original Board VM-IC: Very slow and smooth duty control CM-IC: Moderately fast duty control HM-IC: Very fast and aggressive pulse-density control Variable switching frequency: work harder only when needed. Per cycle averaged duty 1 m. F COUT 20160426 22

Large Signal Load Transient Response: Stretched CM-IC CM-Stretch Original Stretching loop bandwidth CM-IC*1 (condition here) Responding faster than HM-IC, Recovery time ~ 4 -8 us COUT amount is advantage on CM 2 (unstable, choose to use Tantalum cap) (burning more power, with high fsw) 20160426 23

Large Signal Load Transient Response: HM-IC Bigger COUT HM with 200 u. F Original HM-IC with 200 u. F to compare with CMIC’s COUT benefit Responding faster than CM-IC, but slower than original HM-IC Less ringing, very good behavior! 20160426 24

Apple-to-apple comparison LINE TRANSIENT RESPONSE 20160426 25

Line Transient Response VM-IC: Smooth waveforms Recovery time ~ 40 us CM-IC: Edge of stability (0. 1 u. F input cap) Recovery time ~ 80 us HM-IC: Perfect response Recovery time (can’t observe glitches) 20160426 26

Apple-to-apple comparison LOOP STABILITY / SMALL SIGNAL STUDY VOLTAGE-MODE 20160426 27

Stability (Bode/ZOUT): VM-IC Compared with CM-IC (shown later), very consistent gain & phase curves over VIN and IOUT. Harder to design type-III compensation but consistent response once it’s designed. 20160426 See next slide for phase margin values 28

Stability (PM, Small-Signal Transient): VM-IC Condition 5 Vin-5 A 5 Vin-500 m. A 5 Vin-50 m. A 12 Vin-500 m. A 12 Vin-50 m. A Phase Margin from Output from Bode-plot Impedance 51. 3 at 65. 0 k. Hz 61. 0 at 86. 5 k. Hz 54. 0 at 67. 3 k. Hz 66. 1 at 75. 8 k. Hz 50. 3 at 67. 3 k. Hz 62. 5 at 82. 0 k. Hz 49. 5 at 72. 3 k. Hz 58. 4 at 92. 2 k. Hz 57. 6 at 72. 3 k. Hz 68. 6 at 112 k. Hz 59. 6 at 72. 3 k. Hz > 71 Table 1, TI Apps Note SLVA 381 B 20160426 10 m. A load transient over DC current load No ringing observed. 29

Apple-to-apple comparison LOOP STABILITY / SMALL SIGNAL STUDY CURRENT-MODE 20160426 30

Stability (Bode/ZOUT): CM-IC Loop response changes a lot by VIN, IOUT. It’s challenging to support wide changing operation range. 20160426 See next slide for phase margin values 31

Stability (PM, Small-Signal Transient): CM-IC Condition 5 Vin-5 A 5 Vin-500 m. A 5 Vin-50 m. A 12 Vin-500 m. A 12 Vin-50 m. A Phase Margin from Output from Bode-plot Impedance 35. 0 at 55. 7 k. Hz 36. 7 at 64. 8 k. Hz 29. 8 at 60. 4 k. Hz 31. 4 at 70. 1 k. Hz 46. 2 at 67. 3 k. Hz 50. 8 at 82. 0 k. Hz 48. 1 at 77. 7 k. Hz 56. 4 at 88. 6 k. Hz 43. 7 at 80. 6 k. Hz 56. 8 at 112 k. Hz 42. 1 at 80. 6 k. Hz 54. 2 at 99. 7 k. Hz Table 1, TI Apps Note SLVA 381 B 20160426 It is optimized at 12 Vin/6 A… 5 Vin/500 m. A is not good… 32

Apple-to-apple comparison LOOP STABILITY / SMALL SIGNAL STUDY HYSTERETIC-MODE 20160426 33

Bode-plot Measurement: Hysteretic-mode Simulation Waveform • Where to cut the loop? Note: HM-IC is a D-CAP 2™ IC doesn’t have VOUT terminal for the ripple injection. Modified ripple injection RC circuit is used which makes tough to take right Bode plot (detail in later slide). 20160426 34

Bode-plot: Hysteretic-mode w/o Compensation • Without any loop compensation (ripple injection) circuit, nor without using high ESR capacitors, a hysteretic mode device has infinite gain until frequency of f. LC, after f. LC, the gain decline with 40 d. B/dec LC filter slope. • In a real application, this gain will be “ 2 x VIN / (smallest perturbation you can inject)”. Without any compensation, it will end up with rail-to-rail (VIN-GND amplitude) oscillation at VIN at worst. 20160426 35

Bode-plot: Hysteretic-mode with a poor Compensation • With a poor compensation, a hysteretic mode device falls into a double-pulse to multiple-pulse situation. • In case of just double-pulse, there’s a high chance of failure to capture this situation if only its output voltage is monitored. Having a double pulse situation only at a worst condition, not a big issue. Missing double pulse situation at a center condition is not good as it gets worse by sweeping conditions. 20160426 Simulation Waveform 36

Stability (Bode/ZOUT): HM-IC is an Eco-mode™ (skip-mode) IC. 50 m. A load data excluded. IOUT change = RLOAD change Smaller RLOAD decreases Q-factor of LC resonant. 20160426 See next slide for phase margin values 37

Stability (PM, Small-Signal Transient): HM-IC Condition 5 Vin-5 A 5 Vin-500 m. A 12 Vin-500 m. A Phase Margin from Output from Bode-plot Impedance 31. 4 at 237 k. Hz 43. 6 at 236 k. Hz 29. 8 at 221 k. Hz 41. 1 at 201 k. Hz 40. 3 at 229 k. Hz 49. 1 at 201 k. Hz 36. 7 at 213 k. Hz 47. 9 at 201 k. Hz Output impedance gives accurate PM result…let’s see Table 1, Because of un-breakable inner-loop, output TI Apps Note impedance (non invasive) measurement gives SLVA 381 B more accurate phase margin here. 20160426 38