Using Real-Time System Design Methods to Integrate SMPS ... · to Integrate SMPS Control Software...
Transcript of Using Real-Time System Design Methods to Integrate SMPS ... · to Integrate SMPS Control Software...
Using Real-Time System Design Methods
to Integrate SMPS Control Software
with Application Software
Avik Juneja, Alexander G. Dean and Subhashish Bhattacharya Dept. of Electrical and Computer Engineering
North Carolina State University
Raleigh, USA
[email protected], [email protected], [email protected]
Abstract – A switch-mode power supply (SMPS) converts
power efficiently between different voltage levels, making power
optimizations through voltage scaling feasible. SMPS controllers
are generally dedicated hardware (analog/digital circuits,
microcontroller (MCU), digital signal processor (DSP)), and so are
expensive to add to very low cost embedded applications.
In this work, we show how to integrate SMPS control software
into the MCU running application software, which reduces system
cost while increasing the design space and flexibility for
developers. Real-time system design methods are employed to
ensure SMPS voltage regulation quality, while retaining the
original embedded application behavior. Our methods apply to a
wide range of software task schedulers, from simple interrupt-
based foreground/background systems to sophisticated
preemptive real-time kernels (RTOS). We demonstrate our
methods on a position-logging embedded system, with multiple
voltage domains controlled in software, resulting in power savings.
Keywords—SMPS, DVS, DPM, RTOS, DC-DC, Power Supply,
Embedded Systems, Firmware, Real-Time.
INTRODUCTION
Consider an embedded system consisting of an MCU and
peripherals as shown in Figure 1. The operating voltage ranges
of some sample components are depicted in Figure 2 with
vertical bars. One approach is to power all peripherals at 3.3 V,
as shown with the yellow horizontal line. However, we can save
significant power by lowering the operating voltages, as P α V2
for switching and resistive loads, and P α V for constant current
loads. For example, we can add two lower voltage domains
(1.7V (blue), 2.7V (green)) to power the peripherals to operate
at lower voltages.
Standard control approaches assume that the SMPS
controller is implemented with digital logic or on a processor
(MCU or DSP) entirely dedicated to that task [1] [2] [3] [4].
This additional hardware increases system cost and size.
We seek to reduce the cost of adding voltage domains by
controlling their buck converters in software in the same MCU
as the application software. To do this safely, we must ensure
that the control software runs without timing interference from
the application software. We use real-time (RT) system design
techniques to tackle this challenge. A preemptive task scheduler
and task priorities set by real-time design methods ensure
proper responsiveness of all tasks without missing deadlines.
This approach does not require allocation of dedicated CPU
time (such as time slicing). Our contribution is a practical
approach to integrate the SMPS control software into the
application MCU while guaranteeing proper timing and
operation of the SMPS control loops.
Figure 1: An embedded system with common peripherals
Figure 2: Operating voltage ranges of peripherals
A. Scope of Applicability
The embedded application design space and workloads
cover a wide spectrum of power and performance. Depending
on the platform used and the computational demands of the
application, the CPU utilization can vary anywhere from below
This work was supported in part by NSF CSR award 1116850. 978-1-4673-7151-3/15/$31.00 ©2015 IEEE
1% to as high has 100%. Similarly, the power variation between
a performance-limited low-cost MCU and a high end floating
point MCU with many peripherals can be quite significant,
from as low as a few milliwatts to hundreds of watts,
respectively. There is more variability in power vs.
performance on these platforms for wide range of workloads,
depending on software and hardware architecture, and
optimizations, such as fixed point math, direct memory access
(DMA), specialized functional units, ADC operation modes,
variable scaling and optimizations, code reuse, modularity etc.
Our methods provide a means to design and tailor an embedded
application for maximum performance benefits with reduced
power and span a wide range of applications, target hardware
and software. We demonstrate a proof of concept using a
sample embedded application with a variety of peripherals.
Our methods offer significant benefits to system designers,
even when using an unexceptional buck converter without
sophisticated circuit or control features. Performance can be
improved further through synchronous switching (for
efficiency) or feed-forward control (for response) and other
methods. We leave these for future work.
B. Paper Overview
Section II introduces our design methods. Section III presents our experimental platform’s hardware and software. Section IV presents experiments and analysis to evaluate various aspects of the resulting system. Section V summarizes our contributions and presents future directions. Section VI discusses related work.
DESIGN METHODS
A. Overview
Figure 3. Buck converter response to step increase in load current
Our design method is based on understanding the voltage and time characteristics of a buck converter’s response to the worst-case combined load transient, shown in Figure 3. We evaluate the impact of delaying the execution of the control loop software, temporarily increasing the controller delay td. We use real-time system design methods to determine the maximum interference timing interference (delay) possible from the application software. We then put these together to determine the impact of integrating the controller with the application software. We first present an overview of the proposed approach.
Evaluate all of the loads on each supply rail to determine the maximum loading and unloading transients imposed by the rail’s peripherals.
Design a buck converter using standard techniques, ensuring that Vout will not fall below the minimum acceptable voltage when using optimal control.
Design a discrete time control law (compensator) for the buck converter.
Use embedded and real-time system design methods to integrate the control loops into the application software. Leverage hardware peripherals for control operations, providing isolation from application software. Employ preemptive, prioritized task scheduling with interrupt service routines (ISRs).
Use a schedulability test to verify all tasks and ISRs will meet their deadlines.
Use real-time worst-case response time analysis methods to determine the maximum blocking time tb based on duration of other ISRs and other non-interruptible code sections in the application.
Derive an equation ΔVpeak(t) to determine maximum output voltage transient, allowing analysis of impact of delayed execution of the control loop.
Evaluate ΔVpeak(tb) to determine if voltage transient is acceptable. If so, design is complete. If not, adjust design, modify code to reduce tb, change scheduling approach. Repeat analysis.
B. Detailed Discussion
We now examine each of these steps in detail.
1. Evaluate Load Current Transients per Supply Rail Each supply rail r powers peripherals in the set P(r).
Determine the maximum worst-case loading transient for each supply rail r based on peripheral p.
∆𝐼𝑚𝑎𝑥(𝑟) = ∑ ∆𝐼𝑙𝑜𝑎𝑑𝑖𝑛𝑔,𝑚𝑎𝑥(𝑝)
∀𝑝∈𝑃(𝑟)
(1)
2. Design Buck Converter Design a buck converter using standard techniques [5]. The
buck converter shown in Figure 6 is targeted due to its ubiquity, simplicity, low cost and extensive analysis [6] [7] [8].
Select the components and operating parameters such that the maximum combined load transients ΔImax(r) for supply rail r will not cause Vout to fall below the minimum acceptable voltage when using optimal control. A buck converter’s output voltage response to a step change in load current has been characterized [6] [8]:
∆𝑉𝑜𝑢𝑡(𝑡) = −𝑟𝐶∆𝐼𝑜𝑢𝑡 − 𝑡∆𝐼𝑜𝑢𝑡𝐶𝑜𝑢𝑡
+ (𝑡 − 𝑡𝑑)𝑚𝑟𝐶 + (𝑡 − 𝑡𝑑)2𝑚
2𝐶𝑜𝑢𝑡 (2)
Here, m is the slope VL/L of the inductor current. For loading transients VL is (Vin - Vout); for unloading transients it is –Vout. The delay from the load transient to the PWM duty cycle update is represented as td.
The time of the peak output voltage deviation occurs after ¼ of a period of the circuit’s natural frequency [7]. Adding in the ripple voltage may shift this time by up to one switching period:
𝑇𝑝𝑒𝑎𝑘,𝑂𝐿 =𝜋√𝐿𝐶𝑜𝑢𝑡
2 ± 𝑇𝑠𝑤 (3)
This time can be treated as the worst case (smallest) scenario, which will increase with inclusion of circuit parasitic values and resistances.
The size of the inductor relative to the critical inductance value (equation (4)) determines whether the output voltage will rise immediately or after an additional delay [6].
𝐿𝑐𝑟𝑖𝑡 =𝑟𝐶𝐶𝑂𝑢𝑡(𝑉𝑖𝑛 − 𝑉𝑜𝑢𝑡)
∆𝐼𝑜𝑢𝑡 (4)
A smaller inductance (L<Lcrit) results in recovery of voltage immediately after td.
𝑇𝑝𝑒𝑎𝑘,𝐶𝐿 = 𝑡𝑑 (5)
Equation (2) then reduces to:
∆𝑉𝑜𝑢𝑡 = −𝑟𝐶∆𝐼𝑜𝑢𝑡 − 𝑡∆𝐼𝑜𝑢𝑡𝐶𝑜𝑢𝑡
(6)
At that time, the output voltage reaches:
∆𝑉𝑝𝑒𝑎𝑘,𝐶𝐿 = −∆𝐼𝑜𝑢𝑡 (𝑡𝑑𝐶𝑜𝑢𝑡
+ 𝑟𝐶) (7)
However, with a larger inductance (L>Lcrit) the peak output voltage deviation occurs later:
𝑇𝑝𝑒𝑎𝑘,𝐶𝐿 = max 𝑡𝑑 , 𝑡𝑑 +𝛥𝐼𝑜𝑢𝑡𝐿
𝑉𝐿− 𝑟𝐶𝐶𝑜𝑢𝑡
(8)
It is straightforward to apply this time to equation (2) to find ΔVpeak,CL for the closed loop case with a large inductor.
3. Design Buck Converter Controller We next derive the characteristic equation of the buck
converter using state space analysis [5] [9]. This determines open-loop performance parameters such as rise time, settling time, over/undershoot and steady state error. The characteristic equation of the buck converter for a constant current load is approximately:
𝑃 = 1 + (𝑅𝑙𝑜𝑠𝑠𝐶 + 𝑟𝐶𝐶)𝑠 + 𝐿𝐶𝑠2 (9)
This includes the effective parasitic resistances for the non-synchronous buck converter (Rloss = rL + rds*D + rf*(1-D)).
Figure 4: Block diagram of the closed loop system
A mathematical closed loop model was derived, as shown in Figure 4, representing the power stage (converter) and the sampling & control stage (MCU).
The compensator transfer function, D(z), is implemented as a discrete time difference equation using direct control law for simplicity [10]:
𝑑[𝑛] = 𝑑[𝑛 − 1] + 0.2033𝑒[𝑛] − 0.175𝑒[𝑛 − 1]; (10)
4. Implement SMPS Control Next we use embedded and real-time system design methods
to implement the control system for each SMPS.
First, design the system using the MCU’s hardware peripherals to isolate SMPS control operations from software scheduling and task blocking as much as possible. Use hardware timers to trigger the ADC sampling (at fc) and the PWM signal generation (at fsw). Use hardware event signals to transfer data or control signals. Use ADC data update features or use direct memory access to transfer ADC results if possible to eliminate the need for ISRs to perform these operations.
Second, employ a preemptive task scheduling approach and leverage ISRs [11] [12] [13]. The task scheduling approach will dictate how to prioritize task execution at run time, eliminating the need for any time-slicing. Any control loop processing which remains after leveraging hardware peripherals should be executed in a high-priority ISR to minimize blocking. However, it is still vulnerable to blocking when higher-priority ISRs execute, if ISRs are not interruptible, or when ISRs are disabled for data atomicity or other reasons.
5. Verify Task Set Schedulability The next step is to verify that the software task set is
schedulable for a given scheduling approach and priority assignment. This means that no task will ever miss its deadline. This analysis starts by describing each software task and ISR in terms of its worst-case execution time, deadline, maximum release rate, and durations of critical sections (e.g. when interrupts or task switching may be disabled). Suitable tests are well-documented, including both exact response-time tests and inexact but safe utilization-bound tests [14] [15] [16] [17].
6. Find Maximum Blocking Time for SMPS Control Tasks The SMPS control tasks will have been assigned high
priorities. Use well-known methods to determine the maximum blocking time tb for each SMPS control task [14] [17] [18]. These methods include blocking from interference from any higher priority tasks and non-preemptable code segments.
7. Derive Equation for ΔVpeak(tbq)
Figure 5. Impact of blocking time on output voltage
Next we extend the load transient response analysis to consider the blocking time and discrete-time nature of the digital control system. Figure 5 shows the impact of blocking the
Vref[n] e[n] Vo(t)
G(s)
TI - DSP
d(t)
1 −
Kpwm e-sTproc Gvd(s) D(z)
e-sTadc Kadc
control loop execution for blocking times tb1 and tb2. The first time tb1 is short enough that the control still has time to reduce ΔVpeak,CL compared with the open-loop case ΔVpeak,OL. The second time tb2 is long enough that the control responds too late and ΔVpeak,CL = ΔVpeak,OL.
The PWM output duty cycle is updated at a rate of fsw, so the possible time delays tbq are quantized as integer multiples of a switching period Tsw. We assume a one-period delay between ADC sampling and PWM duty cycle updating, but this delay could be reduced in future work.
𝒕𝑏𝑞 = 𝑇𝑠𝑤 ⌈𝑡𝑏𝑇𝑠𝑤
+ 1⌉ (11)
We can now determine the maximum voltage impact from blocking for time duration tbq. There are two cases to consider. If tbq < Tpeak,OL, then control system will have an effect on the peak output voltage:
∆𝑉𝑝𝑒𝑎𝑘,𝐶𝐿 = −∆𝐼𝑜𝑢𝑡 (𝑡𝑏𝑞
𝐶𝑜𝑢𝑡+ 𝑟𝐶) (12)
If tbq > Tpeak,OL then the control system’s response will occur after the open-loop peak and will not affect its magnitude:
∆𝑉𝑝𝑒𝑎𝑘,𝑂𝐿 = −∆𝐼𝑜𝑢𝑡 (√𝐿
𝐶𝑜𝑢𝑡) (13)
8. Evaluate Resulting Voltage Transient If output voltage transient ΔVpeak is acceptable, then this
portion of the design process is complete. If not, the design may be adjusted in various ways.
a) Software Changes
Increasing the control loop frequency will reduce td but will also proportionately increase the computational loading on the MCU. Reducing the blocking time will allow the control system to respond fast enough to reduce the peak voltage transient. This may involve raising the SMPS ISR priority, deferring work from within other ISRs to occur in tasks with lower priorities, making long ISRs interruptible, reducing critical section durations, not disabling all interrupts within critical sections, or adding a priority ceiling protocol.
b) Capacitor Changes
The “brute-force” approach is to increase the size of the output capacitor [7]. Increasing Cout by a factor of x will reduce the slope of the output voltage by a factor of 1 𝑥⁄ . It will delay
the open-loop voltage peak by a factor of √𝑥. Finally, it will
reduce the peak voltage by a factor of 1 √𝑥⁄ . However, this increases cost, volume and board area.
Using a capacitor with a lower ESR, rC. will reduce the ∆𝑉𝑝𝑒𝑎𝑘,𝐶𝐿 but also increase the 𝑇𝑝𝑒𝑎𝑘,𝐶𝐿 . For example, a tantalum
capacitor has a much bigger ESR than the ceramic. However, it will also reduce the critical inductance, making the system more likely to have a slower response even with optimal closed-loop control.
c) Inductor Changes
Increasing the inductance increases the delay until the open-loop peak voltage. For a closed-loop system with an
inductor larger than the critical inductance, the larger inductor will increase the delay between the duty cycle update and the ramp-up of the output voltage. In both cases the magnitude of that peak voltage increases.
Reducing the inductance raises the amount of ripple voltage, unless the switching period is also decreased proportionally (which will increase power losses). The maximum inductor current will also increase, potentially raising inductor costs. This may also lead the converter to operate in discontinuous conduction mode, further complicating control and converter design.
d) Operating Voltage Setpoint
Raising the operating voltage will allow the system to tolerate
a larger ΔVpeak. However, this will also raise the system power
dissipation.
SYSTEM DESCRIPTION
We demonstrate our methods on a prototype position-
logging embedded system, shown in Figure 1. This features a
GPS receiver, WiFi interface, LCD and microSD card. The
system has three voltage domains (3.3 V, 2.7 V, 1.7 V) as
shown in Figure 2, each powered by a software-controlled buck
converter with 5V input, fsw = 50 kHz and fc = 50 kHz. The
application MCU is a 16-bit Renesas RL78/G14 running at 32
MHz with 24 KB of RAM and 256 KB of flash ROM [19] [20].
It can operate from 1.6 to 5.5 V. As with many other
inexpensive embedded MCUs, there is no hardware support for
floating point math. Each of the power supply rails is driven by
a simple non-synchronous buck converter, shown in Figure 6.
The SMPS control software was implemented in the RL78
MCU using a standard control law (eqn. (10)) derived using
digital control theory for a PI compensator [1] [2] [5] [8] [21]
[22].
Figure 6: Circuit diagram and prototype buck converters
The overall control sequence is outlined in Figure 7 and is
implemented with both hardware peripherals and software (in
ADC_ISR). The control task was implemented in a high-
priority ISR. We performed various optimizations to reduce the
computation time. Major benefits came from using fixed point
math, and eliminating many ISRs by using the Event Link
Controller to allow peripherals to trigger each other. We
reduced the computation time tcontrol for one SMPS down from
over 50 µs down to only 2.24 µs with the MCU operating at
32MHz. For three domains the total computation time is
5.56µs. Thus, running these domains consumes only 27.8% of
the MCU’s time, leaving most of it free for application
processing.
Figure 7: SMPS control software sequence diagram
EXPERIMENTAL RESULTS
A. System Operation and Voltage Regulation
Figure 8: Blue: Transients on system 3.3V rail from Wifi activity. Red:
Current drawn by Wifi module, measured as voltage across 1Ω series
resistor
We evaluated the performance of the entire system, which
was fully functional. First we examined the supply rails to
verify proper voltage regulation. As shown in Figure 8 and
Figure 9, the buck converter controllers operate correctly,
regulating the output voltage despite large load current
transients from the WiFi module and SD cards, respectively.
The voltage deviations are acceptable.
Figure 9: Blue: Transients on system 2.7V rail as data is written to SD
card. Red: Current drawn by SD card, measured as voltage across 4.4
Ω series resistor
B. Timing Interference Analysis
We measured the impact of timing interference which
delays (blocks) the execution of the SMPS control ISR,
delaying the response to a load transient. Figure 10 shows the
SMPS response (with optimal control) to a step load transient
with increasing blocking times.
Figure 10: Effects on control task blocking on voltage regulation of
SMPS
Figure 11 shows the impact of different sampling
approaches on the dependence of ΔVpeak on tb. Normal sampling
occurs at the negative edge of the PWM cycle. For just-in-time
sampling (JIT), the ADC is sampled as late as possible to be
able to still reload the PWM counter in the current cycle.
Figure 11: Voltage deviation under normal and JIT sampling technique
for various blocking times
C. Energy Efficiency Analysis
Burst mode operation was added to improve buck converter
efficiency by reducing switching losses when operating under
very low load conditions.
Figure 12 shows that when the duty cycle falls below a
threshold, the controller disables SMPS switching operations so
the load circuit is powered by the output capacitor. When the
output voltage falls below a threshold, the control software
enables SMPS switching operartions, allowing the sytem to
charge the output capacitor again.
0
50
100
150
200
0 20 40 60 80 100
∆V
pe
ak_C
L (m
V)
tb (µs)
Normal Sample JIT Sample
Figure 12: Burst mode implementation of SMPS under very low duty
cycle (small loads)
Figure 13 shows how converter efficiency improves
significantly at lower load currents.
Figure 13: Efficiency improvement due to burst mode operation in
light load condition
Figure 14: Increased efficiency of SMPS at small loads using burst
mode implementation
The overall efficiency curve appears in Figure 14 and Figure
15. If better efficiency is needed at higher loads, the converter
can be changed to a synchronous buck converter to reduce the
conductive losses.
Figure 15: Buck converter efficiency under various load conditions
D. Output Capacitor Selection
We evaluated the transient response of the buck converter using
both ceramic and tantalum output capacitors, as shown in
Figure 16. Ceramic capacitors are more expensive but have a
lower ESR which reduces ripple voltage significantly.
However, this reduces the critical inductance, leading to a
slower response to transients even with optimal control [7] [8].
Figure 16: SMPS response to load transient with different output
capacitor technologies
E. System Energy Use
Three 1F ultracapacitors were employed to power the
system. The total energy available is constant for all test cases.
The ultracapacitors were charged for a fixed amount of time to
voltage Vinit and then allowed to discharge by powering the
system. The amount of time, tdischarge, taken by the system to
drop the voltage of the ultracapacitors to Vfinal determines the
rate of energy consumption or the average power, Pavg
consumed by the system, given as:
𝑃𝑎𝑣𝑔 =𝐸𝑡𝑜𝑡𝑎𝑙
𝑡𝑑𝑖𝑠𝑐ℎ𝑎𝑟𝑔𝑒=1
2𝐶𝑉𝑖𝑛𝑖𝑡2 − 𝑉𝑓𝑖𝑛𝑎𝑙
2
𝑡𝑑𝑖𝑠𝑐ℎ𝑎𝑟𝑔𝑒 (14)
The average power reduction in various operating modes of
the embedded application is normalized and depicted in Figure
17 and Figure 18. It can be clearly seen that using an SMPS
with single voltage domain over a linear regulator results in
power savings of up to 59%. Using multiple domains results in
of power savings of 66%. When comparing the use of single vs.
multiple voltage rail SMPS design, up to 26% power savings
can be observed, as depicted in Figure 18. Although these
0.65
0.7
0.75
0.8
0.85
0.9
0 0.002 0.004 0.006 0.008 0.01 0.012
Load Current (A)
Efficiency @ 2.7V
Mr. Buck Burst Mode
0.7
0.8
0.9
1
0.001 0.01 0.1 1
Load Current (A)
Efficiency
V=2.1V V=2.7V V=3.3V
power gains are modest, they were achieved with three non-
synchronous buck converters with limited efficiency.
Converting them to synchronous converters would improve
efficiency significantly with negligible software impact.
Furthermore, load balancing on supply rails can significantly
improve the average power consumption in all modes of
operation. In the current system, most of the load is on 3.3V rail
(89.15%), resulting in little improvement of multi-rail over
single rail SMPS. Depending on low power applications, the
load can be distributed to 1.7V and 2.7V rails.
Figure 17: Comparison of average power between single SMPS vs.
multi-domain SMPS system
Figure 18: Average power of the three converter setups for various
operating modes
CONCLUSIONS AND FUTURE WORK
By using real-time system design methods to integrate
control software for an SMPS with the application software,
this work enables the SMPS designers as well the embedded
system design engineers to take advantage of a simple
switching power converter in exploiting it for either a tight
power management or loosen the control effort in order to
extract best CPU utilization for other tasks in hand. This
significantly empowers low cost system designers to extract
more efficiency with significant flexibility and scalability for
wide range of applications.
Due to the tight coupling of the hardware and software in
this work, there are limitations which need to be considered
during the design process. These include:
SMPS control software must be robust and isolated from
application software to avoid faults such as memory
overflow, peripheral misuse, etc.
Computational loading of the control software will lead
to application scheduling constraints.
Development time will be increased.
Sub-optimal load distribution may result in limited
energy savings.
Future work exists in various directions. Processor
computational loading can be reduced by leveraging additional
hardware peripherals (e.g. DMA controller), applying dead-
band control, and hardware-assisted feed-forward control.
Transient response can be improved using feed-forward
control. Cost and size can be reduced by increasing switching
frequency. Although this present work examines robustness of
the control software with respect to timing interference by the
application software, the SMPS controller’s peripherals and
memory space must also be protected from interference.
RELATED WORK
In the power supply design community, digital control of switching converters has been a very active field [1] [10] [21] [22] [23] [24]. The transient response has been investigated to identify critical relationships between circuit parameters [7] [25]. It is especially important for voltage regulator modules which must handle the large transients of power-hungry high-performance processors [6] [8].
There have been many investigations into the software issues with controlling power converters using MCUs and DSPs [2] [3] [4]. In fact, many MCU and DSP vendors provide working reference designs for such applications. The real-time design community has investigated the implications of jitter on digital controllers for other applications and suggested various solutions [14] [26] [27].
Designers of high-performance processors (e.g. personal computers and cellphones) have long relied on buck converters for dynamic voltage scaling. There are many papers which concentrate on power management at the microprocessor level [28] [29] [30] [31] [32]. These discuss various techniques to avoid voltage fluctuation by use of sense or prediction techniques, such as application interactions, microprocessor architectural changes, hybrid multi-phase converters, etc.
REFERENCES
[1] Y. Duan and H. Jin, "Digital controller design for switchmode power converters," in Applied Power Electronics Conference and Exposition, 1999. APEC '99. Fourteenth Annual, 1999.
[2] M. Khan and C. Manning, "Microprocessor controlled inverter for UPS applications," in Power Electronics and Variable-Speed Drives, Third International Conference, 1988.
[3] M. Khanniche, "An intelligent real-time microcontrolled UPS system," in Power Electronics and Variable-Speed Drives, 1991., Fourth International Conference, 1990.
[4] M. Sorescu, "Real time microprocessor control loop for switched mode power supply," in Semiconductor Conference, 1997. CAS '97 Proceedings., 1997 International, 1997.
0%
20%
40%
60%
80%
100%
Operating Standalone Low Power Sleep
Relative Average Power
Buck-3.3V Buck-3.3/2.7/1.7
0%
20%
40%
60%
80%
100%
Operating Standalone Low Power Sleep
Relative Average Power
Linear 3.3V Buck-3.3V Buck-3.3/2.7/1.7
[5] R. W. Erickson and D. Maksimovic, Fundamentals of Power Electronics, Second Edition, Kluwer Academic Publishers, 2001.
[6] A. Peterchev and S. Sanders, "Load-Line Regulation With Estimated Load-Current Feedforward: Application to Microprocessor Voltage Regulators," Power Electronics, IEEE Transactions, vol. 21, no. 6, pp. 1704-1717, 2006.
[7] R. Redl, B. Erisman and Z. Zansky, "Optimizing the load transient response of the buck converter," in Applied Power Electronics Conference and Exposition, 1998. APEC '98. Conference Proceedings 1998., Thirteenth Annual, 1998.
[8] A. Peterchev, "Digital Control of PWM Converters: Analysis and Application to Voltage Regulation Modules," University of California, Berkeley.
[9] S. Cuk, "Modelling, Analysis and Design of Switching Converters," Ph.D. Thesis, California Institute of Technology, 1976.
[10] M. Hagen and V. Yousefzadeh, "Applying digital technology to PWM control-loop designs," 2008-2009 Power Supply Design Seminar, 2009.
[11] N. Audsley, I. J. Bate and A. Burns, "Putting fixed priority scheduling theory into engineering practice for safety critical applications," in Proceedings of the IEEE Real-Time Technology and Applications Symposium (RTAS), 1996.
[12] W.-C. Lu, J.-W. Hsieh, W.-K. Shih and T.-W. Kuo, "A faster exact schedulability analysis for fixed-priority scheduling," ournal of Systems and Software, vol. 79, no. 12, pp. 1744-1753, December, 2006.
[13] M. Anand and I. Lee, "Robust and Sustainable Analysis of Embedded Software," in ACM SIGPLAN-SIGBED conference on Languages, compilers, and tools for embedded systems, New York, 2008.
[14] R. Bril, J. Lukkien and W. Verhaegh, "Worst-Case Response Time Analysis of Real-Time Tasks under Fixed-Priority Scheduling with Deferred Preemption Revisited," in ECRTS '07. 19th Euromicro Conference on Real-Time Systems, Pisa, 2007.
[15] I.-G. Kim, K.-H. Choi, S.-K. Park and D.-Y. Kim, "Real-time scheduling of tasks that contain the external blocking intervals," in Second International Workshop on Real-Time Computing Systems and Applications, Proceedings, Tokyo, 1995.
[16] T. Martin, "Digital control for switching converters," in Industrial Electronics, 1995. ISIE '95., Proceedings of the IEEE International Symposium on, 1995.
[17] A. Prodic and D. Maksimovic, "Design of a digital PID regulator based on look-up tables for control of high-frequency DC-DC converters," in Computers in Power Electronics, 2002. Proceedings. 2002 IEEE Workshop, 2002.
[18] S. Kapat and P. Krein, "PID controller tuning in a DC-DC converter: A geometric approach for minimum transient recovery time," in Control and Modeling for Power Electronics (COMPEL), 2010 IEEE 12th Workshop, 2002.
[19] A. Prodic, D. Maksimovic and R. Erickson, "Design and implementation of a digital PWM controller for a high-frequency switching DC-DC power converter," in Industrial Electronics Society, 2001. IECON '01. The 27th Annual Conference of the IEEE, 2001.
[20] K. Yao, Y. Ren and F. Lee, "Critical bandwidth for the load transient response of voltage regulator modules," Power Electronics, IEEE Transactions, vol. 19, no. 6, pp. 1454 - 1461, 2004.
[21] A. Cervin, B. Lincoln, J. Eker, K.-E. Arzen, Buttazzo and Giorgio, "The Jitter Margin and Its Application in the Design of Real-Time Control Systems," in IEEE International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA), Goteborg, 2004.
[22] T. Bund, S. Moser, S. Kollman and F. Slomka, "Jitter Considerations for Worst-Case Performance Generation in Digital Controller Design," Cyber-Physical Systems - Enabling Multi-Nature Systems, 2012.
[23] V. Reddi, M. Gupta, G. Holloway, G.-Y. Wei, M. Smith and D. Brooks, "Voltage emergency prediction: Using signatures to reduce operating margins," in High Performance Computer Architecture, 2009. HPCA 2009. IEEE 15th International Symposium, 2009.
[24] R. Joseph, D. Brooks and M. Martonosi, "Control techniques to eliminate voltage emergencies in high performance processors," in High-Performance Computer Architecture, 2003. HPCA-9 2003. Proceedings. The Ninth International Symposium, 2003.
[25] W. Kim, D. Brooks and G.-Y. Wei, "A Fully-Integrated 3-Level DC-DC Converter for Nanosecond-Scale DVFS," Solid-State Circuits, IEEE Journal, vol. 47, no. 1, pp. 206 - 219, 2012.
[26] E. Grochowski, D. Ayers and V. Tiwari, "Microarchitectural simulation and control of di/dt-induced power supply voltage variation," in High-Performance Computer Architecture, 2002. Proceedings. Eighth International Symposium, 2002.
[27] S. Pontarelli, M. Ottavi, A. Salsano and K. Zarrineh, "Feedback based droop mitigation," in Design, Automation & Test in Europe Conference & Exhibition, 2011.
[28] K. Yao, Y. Ren, J. Sun, L. Kisun, X. Ming, J. Zhou and F. C. Lee, "Adaptive Voltage Position Design for Voltage Regulators," in Applied Power Electronics Conference and Exposition, APEC '04, Nineteenth Annual IEEE, 2004.
[29] S. Sachidananda and A. Dean, "Scheduling Swith-Mode Power Supply Noise for Real-Time Systems," in FREEDM Systems Center Second Annual Conference, Tallahassee, Florida, 2010.
[30] S. Sachidananda and A. G. Dean, "EMI-and Energy -Aware Scheduling of Switching Power Supplies in Hard Real-Time Embedded Systems," in Real-Time and Embedded Technology and Applications Symposuim (RTAS), 17th IEEE, 2011.
[31] N. Rehman, A. Parayandeh, K. Wang and A. Prodic, "Multimode digital SMPS controller IC for low-power management," in International Symposium on Circuits and Systems, ISCAS, 2006.
[32] J. A. Paradiso and T. Starner, "Energy scavenging for mobile and wireless electronics," in Pervasive Computing, 2005.
[33] R. Middlebrook and S. Cuk, "A General Unified Approach to Modelling Switching-Converter Power Stages," Internation Journal of Electronics, vol. 42, pp. 521-550, 1977.
[34] J. R. Farmer, "A comparison of power harvesting techniques and related energy storage issues," Master's Thesis, Virginia Polytechnic Institute and State University, May 2007.
[35] J. R. Farmer, "A Comparison of harvesting techniques and related energy storage issues.," Master's Thesis, Virginia Polytechnic Institute and State University, May 2007.
[36] S. Cuk, "Modelling, Analysis and Design of Switching Converters," PhD Thesis, California Institute of Technology, 1976.
[37] J. Alimeling and W. Hammer, "PLECS-Piece-wise linear electrical circuit simulation for Simulink," in Power Electronics and Drive Systems (PEDS), Proceedings of the IEEE 1999 International Conference, 1999.