A Bipolar-Output Converter with an Adaptive Ramp Generator and PWM Control for OLED Displays

Abstract

Traditional power management for organic electroluminescent (OLED) displays is performed by two different converters, providing both positive and negative power rails and thus raising fabrication costs due to the usage of two inductors, in addition to efficiency being dependent on one of the two converters while also suffering from high-voltage stress. This paper introduces a bipolar converter that only uses a single inductor to generate both power rails for OLED displays, along with two additional flying capacitors to reduce inductor currents and voltage stress placed on the power switches. The proposed design implemented in the 180 nm CMOS process occupies a chip area of 0.98 mm² and achieves a peak efficiency of 93% at 350 mA load current. Furthermore, the ripple voltage is greatly reduced using both common-mode and differential-mode feedback loops in the PWM control scheme, with a maximum ripple of 20 mV across a 100–500 mA current range.

1. Introduction

With increasing demand for larger display sizes and improved quality, OLED displays have gained a reputation as a cost-effective solution in recent years. As usual, a significant portion of the power consumed by electronic devices is used by the displays, making compact, highly efficient DC-DC converter solutions critical for IoTs and wearable devices.

Traditional OLED display power supplies, although practical and straightforward, struggle with on-chip integration due to two bulky inductors since they consist of two different power converters, a boost and an inverting buck–boost converter, to generate both positive and negative voltage. The simplicity of conventional OLED display power supplies comes at the cost of enormous inductor sizes. At the same time, their performance is tied to the efficiency of their internal buck–boost converter, which also suffers from high-voltage stress on power switches. This combination of downsides has driven up fabrication costs and limited circuit performance [1,2,3].

Conventional single-inductor bipolar-output (SIBO) converters have been widely adopted for OLED applications due to their reduced component count; however, they suffer from several inherent limitations, including cross-regulation between positive and negative outputs, increased inductor current ripple under unbalanced loads, and limited efficiency at light-load conditions. These issues have been reported in prior works [4,5,6].

Alternative solutions to SIBO converters include dual-inductor, dual-output converters [4] and switched-capacitor-based bipolar converters [7]. While these approaches can mitigate cross-regulation or improve efficiency under certain conditions, they generally incur higher silicon area, increased external component count, or degraded transient performance. In particular, prior flying-capacitor-assisted SIBO architectures have typically incorporated only a single capacitor [8,9,10]. While effective in reducing switch voltage stress and inductor current to some degree, they leave room for further optimization of inductor current ripple and conduction loss, especially under significant load imbalance. The selection of power switches—using large-sized NMOS transistors for switches connected to lower-voltage nodes (like S₂, S₃, and S₄) to minimize on-resistance R_ON, and PMOS transistors for switches connected to higher-voltage rails (like S₁, S₅, and S₆)—is critical for reducing conduction loss and managing voltage stress across different phases of operation.

Recent hybrid bipolar-output converter designs have favored single-inductor bipolar-output (SIBO) topologies, which now require only a single inductor for both power rails, significantly reducing production costs and product size [4,11,12]. However, they still possess drawbacks. First, the voltage stress on power switches is often greater than 5 V, requiring high-voltage devices. Second, instead of two conducting paths for each inductor, there is now only a single current path; consequently, the inductor current increases, and a larger inductor with a lower DC resistance is needed to avoid conduction loss, ensuring high current and optimal inductance [10,11]. To mitigate these problems, the designs in [12,13,14] have added a flying capacitor, effectively dividing the conduction path into two, charging both the inductor and the flying capacitor, reducing inductor current while simultaneously sharing a voltage of V_IN with the flying capacitor, thus decreasing the |V_IN + V_ON| stress placed on power switches in conventional SIBO converters.

The principal novelty of this work is the introduction of a second flying capacitor on the V_OP output line, forming a dual flying-capacitor architecture. This modification, combined with a PWM control scheme featuring an adaptive ramp generator, creates a more effective dynamic feedback loop. This approach achieves a greater reduction in peak inductor current and output-voltage ripple compared to single flying-capacitor SIBO topologies, without the increase in circuit complexity or silicon area. In this paper, we propose a hybrid bipolar-output converter that uses a single inductor and two flying capacitors to reduce voltage stress and inductor current, employing a PWM control scheme with an adaptive ramp generator to form a dynamic feedback loop, thereby minimizing inductor current and output-voltage ripple on both power rails. Implemented in the 180 nm CMOS process, the converter achieves a 93% peak efficiency at a 350 mA load current and an 84% average efficiency over 150–500 mA, with an output-voltage ripple under 20 mV at any given condition and load.

2. Circuit Implementation

Figure 1 illustrates the overall architecture of the designed SIBO converter. The converter generates output voltages V_OP and V_ON of 4.9 V and −4.7 V, respectively, from a 3.7 V battery supply. The core operational principle involves time-multiplexing the single inductor between the positive and negative output rails across the following four distinct switching phases: two main phases Φ₁ and Φ₂ for balanced energy delivery and two auxiliary phases Φ_P and Φ_N for correcting load imbalance. Switches S_1÷3 and S_4÷6 are controlled by a PWM controller, generating two symmetric main phases and two auxiliary phases for balancing the I_OP and I_ON currents. During the main phases, energy is transferred from the input to both outputs in a balanced manner. The auxiliary phases are selectively activated by the control loop to direct energy preferentially to the output rail with higher demand, thereby maintaining voltage regulation under unbalanced load conditions. Large passive components are the inductor L = 2.2 µH and capacitors C_F1 = C_F2 = 4.7 µF; they are all implemented off-chip and connected to the integrated circuit via the PCB after fabrication and packaging. All switches and level shifters (LSs) are powered internally from V_IN, V_ON, V_OP, or the C_F1/C_F2 capacitors, eliminating the need for external supply rails. A dedicated level shifter (LS₂) ensures proper operation of the S₂ gate driver across the voltage swing on capacitor C_F1. Similarly, level shifter LS₃ is designed for S₃ to translate signals from 0 V to the V_ON level, ensuring a sufficient V_GS for switching. In contrast, switches S₄–S₆ are driven by an on-chip LDO generating V_{OP_1.8}, V_{IN_1.8}, and V_DD, while S₁’s driver is supplied directly by V_IN. In the power stage, switches S₂, S₃, and S₄ are implemented with large-sized NMOS transistors to minimize on-resistance, while S₁, S₅, and S₆ use PMOS transistors.

The inductor value L is determined by limiting the inductor current ripple, which is given based on the work in [15]:

$$\Delta {\mathrm{I}}_{\mathrm{L}}={\displaystyle \frac{{\mathrm{V}}_{\mathrm{I}\mathrm{N}}\times \mathrm{D}}{\mathrm{L}\times {\mathrm{f}}_{\mathrm{s}\mathrm{w}}}}.$$

(1)

To ensure continuous conduction and limit the current ripple to approximately 30% of the average inductor current under maximum load conditions, an inductance in the range of 1.8–2.5 µH is required for a 1 MHz switching frequency. Therefore, an inductance of 2.2 µH is selected as a trade-off between current ripple, conduction loss, and transient response. The flying capacitor is chosen based on the voltage ripple caused by charge transfer, which is calculated in [16]:

$$\Delta {\mathrm{V}}_{\mathrm{C}\mathrm{F}}={\displaystyle \frac{{\mathrm{I}}_{\mathrm{O}\mathrm{U}\mathrm{T}}}{{\mathrm{C}}_{\mathrm{F}}\times {\mathrm{f}}_{\mathrm{s}\mathrm{w}}}}.$$

(2)

Under the maximum load current, maintaining the flying-capacitor voltage ripple below a few millivolts requires a minimum capacitance of approximately 4 µF. Consequently, a 4.7 µF flying capacitor is selected to provide sufficient margin against load variation and process mismatch. The switching frequency is set to 1 MHz to balance switching loss, passive component size, and output-voltage ripple. Operating at this frequency enables compact passive components while maintaining high efficiency and low ripple, which are critical for OLED display applications.

The converter operates through four distinct states (two primary and two auxiliary phases), requiring at least two PWM control signals. As shown in

Table 1. Power switch states in each phase.

Switch	Φ1	Φ2	ΦP	ΦN	Switch	Φ1	Φ2	ΦP	ΦN
S1	ON	OFF	ON	OFF	S4	ON	OFF	OFF	ON
S2	ON	OFF	ON	OFF	S5	OFF	ON	ON	OFF
S3	OFF	ON	OFF	ON	S6	ON	OFF	OFF	ON

, switches S₁–S₃ maintain identical or complementary switching states across all four phases, allowing them to be driven by a single PWM signal (PWM₁, duty cycle D₁). The same principle applies to switches S₄–S₆, which are controlled by a second, synchronized PWM signal (PWM₂, duty cycle D₂).

In the designed bipolar-output DC-DC converter for OLED displays, it is ideal for the V_OP and V_ON sources to have perfectly balanced power. The separate current balancing for the V_OP and V_ON outputs underscores the requirement for identical energy delivery. However, load current imbalance can arise from internal circuits drawing different currents from V_OP and V_ON, or from specific application requirements for independent regulation of these rails. For instance, during startup, the output capacitors (C_OP and C_ON) are discharged to 0V and can be charged to different values, leading to a voltage imbalance. If V_OP is lower than |V_ON|, it indicates a greater energy demand for V_ON, resulting in a higher load current for V_ON (I_ON) compared to V_OP (I_OP), and vice versa.

To address this potential imbalance, dedicated auxiliary phases are implemented to transfer energy specifically to V_OP or V_ON. When I_OP > I_ON, indicating a higher energy demand for V_OP, the phase Φ_P is activated, turning on switches S₁, S₂, and S₅ to transfer energy exclusively to V_OP. Conversely, phase Φ_N is activated to transfer energy only to V_ON via switches S₃, S₄, and S₆. These two phases are illustrated in Figure 2c,d.

Figure 3 shows the oscillator block, a ring of cross-coupled inverters M_23,24, controls the charge/discharge cycle of timing capacitor C_O1. Resistor R_O1 sets the discharge rate. The output at node C_LK is a 1 MHz clock signal used to create the control triangle waves, V_RAMP1 and V_RAMP2, which are essential for regulating the converter’s output voltage. A ring of cross-coupled inverter gates charges and discharges capacitor C_O1 in a stable, continuous cycle. The values of C_O1 and R_O1 determine the oscillation frequency. Transistors M₂₃ and M₂₄ function as inverters without a constant-current source, connected in series to minimize noise and achieve the sharpest possible transition slope. The oscillator, supplied by V_DD, was designed with the parameters shown in Figure 3. To achieve a target T_ON period of 210 ns, transistor M₂₄ was sized to have a high on-resistance. This intentionally introduces a turn-on delay, creating a small duty cycle where T_ON is less than the whole clock period.

Figure 4 shows the schematic for our negative voltage reference circuit; the operation is sequenced by three non-overlapping clock signals C_LKP1,2,3 controlling T-gate switches similar to M_28,39. The charge-pump stage creates V_REFN, which is then regulated via feedback through PMOS switches M_52,53 and operational transconductance amplifier OTA₂, in which three control clock signals C_LKP1,2,3 have a 110 ns pulse width and are turned on sequentially in incremental order of their numbering. A default symmetrical negative voltage is generated using the fundamental principle of a capacitor’s two plates. To create a negative voltage, a positive voltage is first stored across a capacitor. This voltage is then inverted to a negative value at the opposite plate via charge transfer. This charging and inverting process requires precisely timed clock signals to control the switching sequence. The circuit employs T-gate switches M28 and M39, which use complementary CMOS pairs to achieve lower on-resistance than conventional switches. Three clock signals C_LKP1,2,3 with the same frequency but a 90-degree phase shift between each other are used to generate a negative voltage V_REFN with minimal ripple. The generated V_REFN is fed through PMOS switches M_52,53 to regulate the V_NFB node. In operation, any variation in V_REFN due to load changes is detected. The operational transconductance amplifier (OTA2) then outputs an error voltage to stabilize V_REFN through this closed-loop feedback mechanism. This ensures V_NFB remains stable and symmetrical to the positive reference voltage V_REFP. V_REFN and V_REFP serve as the converter’s built-in reference voltages. Conventional converters that lack an integrated negative reference generator require an external reference, increasing the off-chip component area and overall system size.

A differential amplifier with four inputs and two outputs, V_PCM and V_NCM, are shown in Figure 5 to perform signal summation and generate the required error voltages. The amplifier core is based on a stacked folded cascode topology, incorporating PMOS M_59,60,61,62 and NMOS M_63,64,65,66 differential input pairs. This configuration enhances precision by applying the input signal to both PMOS and NMOS transistors, thereby extending the input common-mode range and achieving higher gain. The designed amplifier, denoted as A_CM, consumes a low quiescent power of 6.3 µW. It achieves a high gain of over 90 dB with a small bandwidth of 1.3 kHz, which will help limit amplification at high frequencies and instead compare only low-frequency DC signal differences.

As shown in Figure 6, OTA₁ is a single-stage OTA with two inputs, V_IN and V_IP, and one output V_O1. The design in Figure 6a employs both PMOS and NMOS differential input pairs to enhance precision, providing a wide input range and high gain. Switches M_106,107 are biased to serve as active resistors. They create different reference voltages for the folded-cascode transistors M_102,103 and M_101,111 to control the current consumption in the stacked stage. OTA₁ is designed for low power consumption and low bandwidth, as it is intended for integration within a Type II compensator network. OTA₂ utilizes a current-mirror OTA architecture. The schematic in Figure 6b shows the design of OTA₂, where the quiescent current in each branch M_113,114 is set to 3.7 μA to provide sufficient drive strength. Both OTAs are designed to create stable, fast, and precise current-controlled circuits.

In a bipolar-output DC-DC converter, the V_OP and V_ON voltages can become imbalanced due to constantly varying output loads. To mitigate the difference between the I_ON and I_OP currents, a triangle-wave generator with an adaptive pulse width is implemented; it responds to changes in the V_NDM and V_PDM control signals. Figure 7 shows the schematic and design parameters of this adaptive pulse triangle-wave generator. The circuit uses the previously designed OTA₂ as a controlled current source, acting as a small LDO whose output is modulated by the V_NDM and V_PDM voltages. A fixed bias current is provided, ensuring the ramp signal is never completely turned off during startup, even when V_PDM and V_NDM are at the V_SS or V_DD rails. The ramp signal is generated by exploiting the charging and discharging characteristics of capacitors C_A1 and C_A2, which are controlled by the clock signal C_LK from the oscillator block.

In Figure 8a, the schematic of the first level shifter LS1 is shown. It converts an input square wave S_IN with a 1.8 V upper rail and a 0 V lower rail to an output wave whose rails are defined by the applied voltages V_H and V_L, spanning from V_IN down to 0 V. Furthermore, Figure 8b presents the schematic of the second level shifter LS2. While its input is also a 1.8 V/0 V square wave, LS2 is supplied with a negative voltage at V_LN2 and a positive voltage at V_HN2. This is because the LS2 block is designed to drive switches S₂ and S₃ with a negative ON voltage V_ON.

Both the LS1 and LS2 blocks employ a current-mirror structure. When the S_IN signal transitions, the control switches toggle, creating a voltage imbalance. The current-mirror configuration detects this minute change and swiftly generates the corresponding output-voltage levels (V_H/V_L for LS1 and V_HN2/V_LN2 for LS2). To ensure the level-shifted output signal has sufficient current strength to drive the switches S_1÷6, the core inverter stages within the shifter are sized up to increase the charging current.

3. Results

Figure 9 illustrates the layout arrangement of the complete bipolar-output DC-DC converter, including the placement of the power switches S_1÷6. With a chip area of 0.98mm², the converter consists of two main control blocks, namely the DM (differential mode) controller and the CM (common mode) controller, which are placed at the center of the chip in order to deliver control clock signals to other blocks, mainly the level shifters, with as little delay time as possible. The power switches S_1÷6 are placed on the outskirt of the chip for easy access to I/O pads. The layout was verified using Calibre through LVS and DRC checks. Parasitic components in the layout were carefully optimized to ensure high accuracy and reliability during fabrication.

Figure 10a shows that the output voltages require approximately 500 μs to stabilize from startup, reaching steady-state values of V_OP = 4.9 V and V_ON = −4.7 V. The simulation was performed using 0.18 μm technology node with the following parameters: a default positive reference voltage V_REFP = 450 mV, a negative reference voltage V_REFN = −430 mV, an inductor L = 2.2 μH, output capacitors C_OP = C_ON = 10 μF, a switching frequency of 1 MHz, and a supply voltage of 3.7 V. Figure 10b shows the V_PCM and V_NCM signals during the stable operation of the bipolar-output DC-DC converter. At initial startup, the voltage difference between V_FBN/V_FBP and V_REFN/V_REFP is large, causing the A_CM block to act as a comparator. Once the converter stabilizes at the desired output voltages, the V_PCM and V_NCM signals become equal.

Figure 11a shows the waveforms of V_PDM and V_NDM. During the initial startup phase, these signals continuously adjust to balance the I_OP and I_ON currents by controlling the auxiliary phases of the power switches network. Figure 11b illustrates the corresponding modulation of the D₁ and D₂ duty cycles to correct any imbalance between I_OP and I_ON. A disparity between the D₁ and D₂ duty cycles triggers the automatic activation of an auxiliary phase to restore balance. This balancing occurs both at startup and during load transients, such as transitions from high to low load. The modulation is achieved by varying the peak of the triangle wave while keeping the VSUM signal constant, necessitating adjustments to the D₁ and D₂ duty cycles.

Figure 12a shows the waveforms of V_ON, V_OP, and the inductor current under a 300 mA load. The voltage ripple on V_ON and V_OP is 14 mV, and the inductor current ripple is 440 mA. It can be observed that, despite variations in load current, V_ON and V_OP remain stable at approximately −4.7 V and 4.9 V, respectively. Furthermore, a lower load current results in reduced voltage ripple on V_ON and V_OP, as the inductor current ripple decreases. Figure 12b shows the converter’s efficiency as a function of the load current. This result was plotted with all the losses accounted for, including the power consumptions of different blocks (gate driver, level shifter, ….) as well as switching loss and conduction loss inflicted onto the MOSFETs inside the power stage. The peak efficiency is 93%, which occurs at an output current of 0.35 A.

Table 2. Performance comparison.

Items	This Work	[13]	[17]	[18]
Process (μm)	0.18	0.5	0.09	0.35
Topology	SIBO (PWM)	SIBO (PWM)	SIBO (PWM)	SIBO (PWM)
Switching freq. (MHz)	1	1.4	0.5	1
Input voltage (V)	3.7	3.7	3.7	3.7
Output voltage (V)	−4.7	−4.9	−2.5	−4.7
	4.9	4.6	4.8	5.3
IOUT range (mA)	100–500	300	30	30–350
Inductor (μH)	2.2	10	10	10
Capacitor (μF)	10	10	10	10
Flying capacitor	2 × 4.7	4.7	N/A	4.7
Output ripple (mV)	20 (500 mA)	8.5 (30 mA)	17–50 (20 mA)	<30 (30–350 mA)
Power efficiency (%)	93 (350 mA)	94 (100 mA)	89 (10 mA)	88 (100 mA)
Active chip (mm2)	0.98	1.46	N/A	3.68

presents a performance comparison between the proposed converter and similar state-of-the-art SIBO converters. The proposed design, implemented in a 0.18 µm CMOS process, achieves a peak efficiency of 93% at 350 mA, which is competitive with the 94% efficiency of [13] and superior to the 88–89% efficiencies of [17,18]. A key advantage of this work is its significantly wider output current range of 100–500 mA, supporting heavier loads than the compared designs. Furthermore, it accomplishes this with a substantially smaller inductor value of 2.2 µH, a reduction of over 75% compared to the 10 µH required by [13,17,18], which is crucial for system miniaturization. This design also achieves the smallest active chip area of 0.98 mm², demonstrating a more area-efficient integration despite not using the most advanced process node.

4. Conclusions

The designed bipolar-output DC-DC converter, implemented in a 180 nm CMOS process, achieves a peak efficiency of 93%. It provides stable output voltages of V_ON = −4.7 V and V_OP = 4.9 V, effectively supporting a load current range of 0.1–0.5 A. The converter’s primary application is to supply a symmetric voltage for OLED displays, enhancing their luminous efficiency and stability. The topology utilizes two flying capacitors in parallel with the inductor, enabling a significant reduction in the inductor’s size. Despite the use of external flying capacitors and on-chip switches, the converter core occupies a compact area of 0.98 mm². Furthermore, the implemented PWM control scheme ensures a fast transient response to load variations.