Results - International Rectifier

Freescale – T4240 RDB
IRF Proven Power Design with
Freescale’s T-Series QorIQ
For Complete Design Information
http://mypower.irf.com/Freescale
1
Contents & Guidelines of Report
• Results show design of scaling processor cores vs.
power design from 24 Cores to 4 Cores
• IRF and Freescale collaboration of a Power design
to scale from T4 to T1 processors – T4240 used to
scale the design
Core and Memory Power Rails
For T4 Designs:
IR3565BMFS01TRP
For T2/T1 Designs:
• Design was verified with Freescale on the
RDB/QSS/HSSI designs and given to ODM/IDH
partners to seed production level reference designs
• This report highlights the advantages of the Digital
Power design to “Tune” for actual load parameters
and system level processor modes for high
performance scaling.
Core+DDR:
IR36021MFS01TRP
Core Only:
IR36021MFS02TRP
• International Rectifier now offers PRE-CALIBRATED
Multi-phase Power devices optimized for Freescale
T-Series QorIQ Processors to meet the dynamic,
tight design requirements from 10A to 80A for the
core and memory rails.
• All other IR devices in the power solution are
standard (SupIRBucks & FETs) for the peripheral &
I/O rails.
IRF Power on the following:
T4240, T2080, T1042, T1040, more coming…
2
T4240 – VCORE & DDR Sol’n Overview
IR3565B , 4+2-Phase Dual Output Digital Controller
IR3550, 60A PowIRstage
FP1007R3-R15-R Inductor on 4-Phase VCORE (Fsw = 500kHz)
L = 150nH / DCR = 0.29mOhm & Csen = 220nF
Set Rsen = 2.40kOhm (Rsen >= L/DCR/Csen)
Rcs = 3.40kOhm, Rs1_s2 = 2.87kOhm, Ccs = 100pF
Rll ~ 1.0mOhm & AVP BW ~ 192kHz
IHLP-4040-DZ Powdered Iron Inductor on 2-Phase DDR (Fsw = 500kHz)
L = 360nH / DCR = 1.30mOhm & Csen = 220nF
Set Rsen = 1.30kOhm (Rsen >= L/DCR/Csen)
Rcs = 2.43kOhm, Rs1_s2 = 1.80kOhm, Ccs = 120pF
Rll ~ 2.5mOhm & AVP BW ~ 200kHz
3
T4240 – VCORE (Loop 1) – NTC & LL Comp
4
T4240 – DDR (Loop 2) – NTC & LL Comp
5
iScale, iGain, and Offsets – DC Calibration
Loop_1_iscale = 581 dec = 0x0245 ~ 92.87% of nominal Rll scaling
Loop_1_i_gain = 1 dec = 0x01 ~ 4x AFE Gain
1.0mOhm * 0.9287 = 0.9287mOhm Actual Rll from RCSP/RCSM Circuit
(DCR is acting ~ 7.7% high… 0.29mOHm * 1.077 ~ 0.312mOhm effective DCR)
Offsets
0.0A Offset Req’d (0.5A LSB optional +3.5A/-4.0A)
0.0mV Offset Req’d (5mV LSB optional +40mV/-35mV)
Loop_2_iscale = 215 dec = 0x00D7 ~ 85.92% of nominal Rll scaling
Loop_2_i_gain = 0 dec = 0x00 ~ 2x AFE Gain
2.5mOhm * 0.8592 = 2.148 mOhm Actual Rll from RCSP/RCSM Circuit
(DCR is acting ~ 16.4% high… 1.30mOHm * 1.164 ~ 1.513mOhm effective DCR)
Offsets
0.0A Offset Req’d (0.5A LSB optional +3.5A/-4.0A)
0.0mV Offset Req’d (5mV LSB optional +40mV/-35mV)
6
T4240 – VCORE – Load Model
7
T4240 – VCORE – PID Tuning & Bode Model
8
T4240 – VCORE – Actual BODE – Stable PIDs
9
T4240 – VCORE – Actual BODE – Unstable PIDs
10
T4240 – Vcore – PID Tuning & BODE Analysis
VCORE STABLE SETTINGS (Bode Optimized – used these!)
Kp = 34, Ki = 30, Kd = 48, LPF1 = 4, LPF2 = 9
Modeled Data
PM = 35.1deg
GM = 22.5dB
BW = 36.5kHz
Real Data
77.6deg
20.8dB
71.6kHz
VCORE UNSTABLE SETTINGS (Response Optimized)
Although the Vout response looked marginally better, overall stability and robustness /
loop-stability is a concern.
Loop stability calls for Loop BW << 1/5 Fsw.
Fsw = 500kHz. So, our Loop BW should be << 100kHz.
Kp = 36, Ki = 32, Kd = 53, LPF1 = 4, LPF2 = 9
Modeled Data
Real Data
PM = 62.1deg
78.8deg
GM = 15.5dB
10.3dB
BW = 50.0kHz
149.6kHz
11
ORIGINAL PIDs – 16A Loadstep (37.5A – 53.5A)
24 COREs using Freescale Dhrystone Code
Vmin = 1.026V
Vmax = 1.084V
Droop ~ 24.0mV
Overshoot ~ 34.0mV
TOB = +/- 30.0mV
TOB = +/- 30.0mV
Margin = 6.0mV
Margin = (-)4.0mV
12
STABLE PIDs – 16A Loadstep (37.5A – 53.5A)
24 COREs using Freescale Dhrystone Code
Vmin = 1.0250V
Vmin = 1.0255V
Droop ~ 25.0mV
Droop ~ 24.5mV
TOB = +/- 30.0mV
TOB = +/- 30.0mV
Margin = 5.0mV
Margin = 5.5mV
13
UNSTABLE PIDs – 16A Loadstep (37.5A – 53.5A)
24 COREs using Freescale Dhrystone Code
Vmin = 1.0300V
Vmin = 1.0322V
Droop ~ 20.0mV
Droop ~ 17.8mV
TOB = +/- 30.0mV
TOB = +/- 30.0mV
Margin = 10.0mV
Margin = 12.2mV
14
STABLE PIDs – 16A Load-Release (53.5A – 37.5A)
24 COREs using Freescale Dhrystone Code
Vmax = 1.0796V
Vmax = 1.080V
Overshoot ~ 29.6mV
Overshoot ~ 30.0mV
TOB = +/- 30.0mV
TOB = +/- 30.0mV
Margin = 0.4mV
Margin = 0.0mV
15
UNSTABLE PIDs – 16A Load-Release (53.5A – 37.5A)
24 COREs using Freescale Dhrystone Code
Vmax = 1.0760V
Vmax = 1.0788V
Overshoot ~ 26.0mV
Overshoot ~ 28.8mV
TOB = +/- 30.0mV
TOB = +/- 30.0mV
Margin = 4.0mV
Margin = 1.2mV
16
STABLE PIDs – 5A Loadstep (37.5A – 42.5A)
8 COREs using Freescale Dhrystone Code
Vmin = 1.0409V
Vmin = 1.0402V
Droop ~ 9.1mV
Droop ~ 9.8mV
TOB = +/- 30.0mV
TOB = +/- 30.0mV
Margin = 20.9mV
Margin = 20.2mV
17
STABLE PIDs – 5A Load Release (42.5A – 37.5A)
8 COREs using Freescale Dhrystone Code
Vmax = 1.0638V
Vmax = 1.0629V
Overshoot ~ 13.8mV
Overshoot ~ 12.9mV
TOB = +/- 30.0mV
TOB = +/- 30.0mV
Margin = 16.2mV
Margin = 17.1mV
18
T4240 – DDR Load Model
19
T4240 – DDR PID Tuning & Bode Model
20
T4240 – DDR Actual BODE Stable (used!)
21
T4240 – DDR – PID Tuning & BODE Analysis
DDR STABLE SETTINGS (Bode Optimized – used these!)
Kp = 35, Ki = 32, Kd = 52, LPF1 = 4, LPF2 = 9
Modeled Data
PM = 56.1deg
Real Data
61.7deg
GM = 17.3dB
BW = 41.0kHz
24.1dB
32.4kHz
22