A High Voltage Dickson Charge Pump in SO1 CMOS Mohammad R. Hoque, Ty McNutt, Jimmy Zhang, Alan Mantooth, Mohammad Mojarradi* University of Arkansas, 32 17 Bell Engineering Center, Fayetteville, AR Jet Propulsion As, Cali*nia Institute of Technology, 4800 Oak Grove m, Pasadena, CA Drive lab&I U Abstract An improved charge pumpt that utilizes a MOSFET body diode as a charge transfer switch is discussed. The body diode is characterized and a body diode model is developed for simulating the charge pump circuit. An increase in voltage pumping gain for a silicon-on-insulator (SOI) Dickson charge pump is demonstrated when compared with a traditional bulk CMOS Dickson charge pump. A 6-stage Dickson charge pump was designed to produce a 20 V output from a 3.3 V supply, using a 4 MHz, two-phase non-overlapping clock signal driving the charge pump. The design was fabricated in a 0.35 pm partially depleted SO1 CMOS process. An efficiency of 72% is achieved at a load current of approximately20 pA. I. Introduction As the desire for unmanned spacecraft and interplanetary exploration increases and the enabling technology comes to fruition, MEMS devices such as microgyroscopes (for navigation), microseismometers (for detecting seismic activity), micrometeorological equipment (such as a microhygrometer for measuring humidity), mass spectrometers (for identification and classification of chemical species), and micropropulsion engines become invaluable in the cost-effective push in space related research [I]. MEMS devices such as these require a range of DC bias voltages, ftom 15 V to well over 60 V, to perform the necessary environmental interactions [ 1-31. Among the different on-chip, high-voltage generators, the Dickson charge pump is a classic solution [4]. However, the pumping efficiency of the bulk-CMOS Dickson charge pump is degraded due to the threshold voltage drop of the charge transfer switch (CTS) transistors, and the voltage gain of the pumping stages closer to the output is further degraded by the increase in threshold voltage due to the body effect of these MOSFET switches [5]. The utilization of SO1 MOSFET body diodes in place of the typical bulk MOSFET switches reduces the voltage drop across the switches significantly. The goal of this paper is to present improvements achievable in an SO1 Dickson charge pump circuit as compared to a bulk CMOS realization. While no comparisons to other charge pump circuit topologies are drawn, the improvements should be equally applicable to many other bulk CMOS DC/DC converter circuits when implemented in SOI. This work was supported by a NASA Research Award No. NASA- 215. 11. Background Two popular topologies for stepping up DC voltages are the charge pump DC/DC converter [4] and the boost switching DC/DC converter [6]. Each of these D O C converter topologies offers performance advantages and disadvantages. In the past, the boost converter has been the predominantly used converter. The boost converter can provide much more power than the charge pump and is more than 90% efficient, but the boost converter requires large inductors. For many on-chip applications it is desirable to avoid the use of inductors because of their size, non-linearity, and interference. It is for this reason that charge pumps have been so widely used in integrated circuit (IC) applications. Charge pumps are designed using MOSFETs or diodes as switches, and they utilize energy-transfer capacitors instead of inductors, thus enabling them to be readily implemented on-chip . Charge pump switches can be diodes, diode-connected MOS transistors, or MOS body diodes. While bulk MOSFET transistors have a significant parasitic drain capacitance and latch-up issues, an SO1 process can be used to overcome these limitations [7-91. Other compelling advantages of SO1 technologies are excellent tolerance of transient radiation effects and superior thermal properties [8,10]. In bulk CMOS circuits, the p-n junction between the substrate and well forms a “parasitic” diode. The breakdown voltage of this diode is a limiting factor for pumping high voltages if this type of transistor is used in a charge pump. If a potential in the charge pump is reached which is greater than the value of the breakdown voltage, the circuit will cease to operate properly, if at all. However, in an SO1 process this substrate-well diode does not exist, since the transistors are isolated from each other and built upon an insulating oxide layer. With proper circuit topology selection, this affords the opportunity to pump to much higher voltages on-chip, where the limitations will now be oxide breakdowns for capacitors and/or the buried oxide layer [ 111. 111. SO1 MOSFET Body-Diode Switch In an SO1 process, each MOSFET body is isolated from other neighboring transistors due to the buried oxide (BOX) layer. The cross-section of an NMOS SO1 transistor is shown in Fig. 1. The two p-n junctions of this structure form a back- to-back diode configuration as indicated in the drawing [12]. The MOSFET CTS connection used in the SO1 Dickson charge pump developed in this work is implemented by connecting the gate (G), drain (D), and body (B) together. In this implementation, the floating-body of the NMOS is connected to the drain, thus shorting the junction from the p- type body to the n-type drain. Now, only one diode exists