A Survey of Circuit Innovations in Ferroelectric Random-Access Memories Ali Sheikholeslami, MEMBER, IEEE, AND P. Glenn Gulak, SENIOR MEMBER, IEEE This paper surveys circuit innovations in ferroelectric memo- and low power consumption and the emergence of new ries at three circuit levels: memory cell, sensing, and architecture. applications such as contactless smart cards and digital A ferroelectric memory cell consists of at least one ferroelectric ca- cameras. pacitor, where binary data are stored, and one or two transistors that either allow access to the capacitor or amplify its content for Table 1 compares ferroelectric memories with elec- a read operation. Once a cell is accessed for a read operation, its trically erasable and programmable read-only memories data are presented in the form of an analog signal to a sense ampli- (EEPROM’s) and Flash memories, two types of floating-gate fier, where it is compared against a reference voltage to determine memories, in terms of density, read-access time, write-access its logic level. time, and the energy consumed in a 32-bit read/write. En- The circuit techniques used to generate the reference voltage must be robust to semiconductor processing variations across the chip joying a mature process technology, EEPROM’s and Flash and the device imperfections of ferroelectric capacitors. We review memories [1], [2] are superior to ferroelectric memories in six methods of generating a reference voltage, two being presented terms of density. Also, they require less power compared to for the first time in this paper. These methods are discussed and ferroelectric memories for read operations, a factor that will evaluated in terms of their accuracy, area overhead, and sensing keep them popular in applications that demand numerous complexity. Ferroelectric memories share architectural features such as memory reads but only occasional memory writes. An addressing schemes and input/output circuitry with other types example of such applications is an identity card where an of random-access memories such as dynamic random-access identity code is programmed into the memory once but read memories. However, they have distinct features with respect to many times afterwards. accessing the stored data, sensing, and overall circuit topology. We Ferroelectric memories, on the other hand, are superior review nine different architectures for ferroelectric memories and discuss them in terms of speed, density, and power consumption. to EEPROM’s and Flash memories in terms of write-access time and overall power consumption, and hence target ap- Keywords—Ferroelectric memory, memory circuit design, non- plications where a nonvolatile memory is required with such volatile memory. features. Two examples of such applications are contactless smart cards and digital cameras. Contactless smart cards re- I. INTRODUCTION quire nonvolatile memories with low power consumption, as For the last three decades, floating-gate memories have they use only electromagnetic coupling to power up the elec- been the dominant class of nonvolatile memories in ap- tronic chips on the card. Digital cameras require both low plications ranging from personal computers to consumer power consumption and fast frequent writes in order to store electronics. In recent years, however, ferroelectric memories and restore an entire image into the memory in less than 0.1 s. have received more research attention as evidenced in recent Another advantage of ferroelectric memories over inventions in this field. In the past three years, for example, EEPROM’s and Flash memories is that they can be easily there have been more than 320 patents granted by the U.S. embedded as part of a larger integrated circuit to provide patent office. More than 120 of these inventions have been system-on-a-chip solutions to various applications [5], [6]. granted during the past year alone. This increased level of Future personal wireless connectivity applications that are activity is being driven by two motives: superior features battery driven, such as third-generation cellular phones and of ferroelectric memories such as short programming time personal digital assistants, will demand large amounts (mul- tiple megabytes) of nonvolatile storage to retain accessed Internet Web pages, containing compressed video, voice, Manuscript received July 14, 1999; revised February 23, 2000. This work and data. The density and energy efficiency of writing data was supported in part by Fujitsu, Kawasaki, Japan; in part by Nortel Net- works, Canada; and in part by the Natural Sciences and Engineering Re- to memory would seem to indicate that ferroelectric memory search Council of Canada. will play a major role in these types of consumer products. The authors are with the Department of Electrical and Computer En- As shown in Fig. 1, a ferroelectric memory technology gineering, University of Toronto, Toronto, ON M5S 3G4 Canada (e-mail: [email protected]; [email protected]). consists of a CMOS technology with added layers on top for Publisher Item Identifier S 0018-9219(00)04568-0. ferroelectric capacitors. Therefore, by masking parts of the 0018–9219/00$10.00 © 2000 IEEE PROCEEDINGS OF THE IEEE, VOL. 88, NO. 5, MAY 2000 667 Table 1 Nonvolatile Memories and Their Typical Features * Data given for the basic memory cell and may be different when designed into an integrated circuit. y This number is based on nonstacked ferroelectric capacitor technology. More advanced fabrica- tion technologies (i.e., with stacked capacitor [3]) offer comparable area/cell to that of EEPROM. Fig. 1. Ferroelectric capacitor layers (two electrodes and a thin film of ferroelectric material) on top of a conventional CMOS process. design that are not using ferroelectric capacitors, CMOS dig- ital and analog circuits can be integrated together with ferro- electric memories, all in the same chip. Fig. 2 illustrates the cross section of a ferroelectric memory technology [3], [4] that allows the ferroelectric capacitors to sit directly on top of the transistors by means of stacked vias, hence reducing cell area. Research on ferroelectric memories is proceeding on three fronts: material processing [7]–[14], modeling [15]–[17], and circuit design. On the material front, we offer a brief review of the historical evolution of ferroelectric materials for ferroelectric memories and compare them with fer- romagnetic memories. This is presented in Section II of this paper, along with a brief review of the fundamental characteristics of ferroelectric materials and their usage as a medium for nonvolatile storage of binary data. Accurate modeling of ferroelectric capacitors is essential to circuit simulation and design of ferroelectric memories. A Fig. 2. Cross section of a ferroelectric memory technology [3] circuit-based model, for example, can be integrated into cir- that uses three metal layers and allows stacked vias to minimize the memory cell area. cuit simulation tools such as HSPICE [18] to simulate var- ious sections of a ferroelectric memory. We refer the inter- ested readers to [15] for a recent survey of behavioral mod- ferroelectric memories. Section III of this paper presents eling of ferroelectric capacitors and to [16] for a recent mod- an overview of the basic circuit operations of ferroelectric eling effort by authors of this paper. memories and their terminology. Section IV of this paper Our main focus in this paper is on innovative circuit presents circuit innovations that relate to reference voltage techniques. We survey research efforts on this front as generation techniques. Six different methods are reviewed it relates to circuit innovations for higher performance and compared in terms of their sensing speed, accuracy, area 668 PROCEEDINGS OF THE IEEE, VOL. 88, NO. 5, MAY 2000 Fig. 4. Hysteresis loop characteristic of a ferromagnetic core of Fig. 3. Fig. 3. Two-dimensional array of ferromagnetic cores. Each core is accessed by a simultaneous current pulse on an x-access and y-access wire [19]. The read operation as explained above is destructive since a “0” is written to any core that is accessed for a read. The original data, however, are saved at the sense amplifier and overhead, and sensing complexity. Two of these methods are can be restored back into the accessed core. In other words, presented for the first time in this paper. Section V reviews a read access is only complete after the second write that circuit innovations at the architectural level and highlights restores the original data. their advantages and disadvantages over conventional archi- The success of the core memory was due to its simple ar- tectures. Section VI offers a brief overview of technology chitecture and a square-like hysteresis loop characteristic of trends. Last, Section VII presents our conclusions on the the core, as shown in Fig. 4. The architecture allows storing circuit design aspects of ferroelectric memories. a memory bit at every intersection of the access wires, re- sulting in a scalable, relatively dense array of cells. This ar- II. BACKGROUND chitecture, with a few modifications that make it more suit- The underlying principles of operation for ferroelectric ca- able for higher capacities, has been adopted by many gener- pacitors and ferromagnetic cores are similar. In this section, ations since the core memory. In fact, most semiconductor we present a brief review of the principles of operation of fer- memories today, including dynamic random-access memo- romagnetic memories and their counterparts in ferroelectric ries (DRAM’s), EEPROM’s, and ferroelectric random-ac- memories. cess memories (FRAM’s), use a very similar row–column architecture. A. Ferromagnetic Cores A square hysteresis loop, as shown in Fig. 4, displays the Prior to the 1950’s, ferromagnetic memories (also known magnetic flux of a core in terms of the electric current as core memories) were the only type of random-access, non- passing through it. The two points corresponding to a zero volatile memories [19]. A core memory, as shown in Fig. 3, current are marked to represent the two nonvolatile mag- consists of a regular array of tiny magnetic cores that can be netic states of the core.