Advanced Memory Technology

1 Terabit on a Chip – New Memory Technology Rises to Challenge NAND Flash

Memory devices without transistors but based on the migration of oxygen ions under an electrical field promise low-cost memory speeds and densities well beyond what is now available.


  • Page 1 of 1
    Bookmark and Share

Article Media

A new nonvolatile memory technology is gearing up to challenge high-density NAND Flash memory in both commodity mobile devices as well as in enterprise storage. It is being developed by Unity Semiconductor and is called conductive metal oxide (CMOx). It is built around memory cells that are connected in a cross-point array and are not based on transistors, but rather a technology that uses the movement of ionic charge under control of an electric field. The immediate advantages are that in the initial implementation, four layers of 2-bit multilevel cells make it possible for the die to be almost three times as dense as a NAND die of the same size, and it is based on a cell size that starts smaller than NAND and will continue to be reduced in size.

The basic technical breakthrough of CMOx is that bits are stored in terms of moving ions rather than electric charges. A CMOx cell consists of a two-terminal memory element between two electrodes (Figure 1). One of these is a thin insulating metal oxide and the other is a conductive metal oxide. The exact composition of these two materials is Unity’s key intellectual property. The conductive metal oxide (CMO) acts as a reservoir of oxygen ions that are pulled into and out of the insulating metal oxide (IMO) under control of an electrical field. The atoms themselves obviously do not move, but the transfer of electrons causes the net charge on the atoms to shift. The shift of ions back and forth from the IMO changes the resistivity of the cell. The IMO acts as a tunnel barrier to limit current flow through the device.

Figure 1
In its unprogrammed state, the CMOx cell is passing a relatively high read current due to the fact that there are no oxygen ions in the tunnel barrier insulating the metal oxide layer. When a programming voltage totaling 2.5V is applied, ions will be driven into the IMO layer increasing resistivity and resulting in a lower read current. To reprogram (erase) the opposite programming voltage is applied.

To read the cell, a small voltage of 1.0V is applied across it and the current flow is measured to determine the “data” content of the cell. In the erased state, oxygen ions are distributed in the CMO layer and the IMO layer is relatively free of oxygen ions. In this state, the tunnel current is relatively high so the read current through the cell is high. To program the cell, an electric field of a higher voltage (2.5V) is put across the cell causing a migration of oxygen ions into the IMO with the result of lowering the read current through the cell. One of the tricks of this technology is to be sure that the read voltage does not cause a migration of oxygen ions but that such migration will occur under the program/erase voltage of 2.5V. Erasing the cell simply involves putting a reverse 2.5V field across it driving the oxygen ions back into the CMO layer.

The effects of these voltages on the migration of ions and hence on the amount of read current are quite nonlinear, and Unity has discovered that by changing the program/erase voltage, they can control the number of ions that migrate and thus can establish multiple read current levels at the same 1.0V read voltage. Since the effects are nonlinear, Unity has been able to determine thresholds on the current curve that reliably represent two levels, resulting in the ability to store two bits per cell. NAND is currently able to store up to three bits per cell, but COE Dave Eggleston indicates that given the other density and performance advantages of CMOx, he is currently happy with two bits per cell. 

That is not an idle decision given the fact that the initial commercial implementation of CMOx memory cells will be in a four-layer array (Figure 2) with FET decode logic underlying the cell arrays. The CMOx memory is not byte addressable like DRAM and PCM, but rather data is written and read as 4 Kbyte pages streaming at 200 Mbyte/s for write and 500 Mbyte/s for read. At the projected speeds and densities, Unity is developing a 1 Terabit chip arranged in 2 Mbyte tiles in a hierarchical architecture shown in Figure 3. With the MLC capability of two bits per cell, the capacity adds up to a Terabit on a single chip.

Figure 2
The first commercial implementation of CMOx memory foresees a four-layer structure with each cell capable of storing two bits. CMOS decode logic will underlay the cross-point memory array.

Figure 3
The first CMOx storage chip will be a one Terabit device built up of a hierarchical architecture of memory tiles, bricks and planes. 4 Kb page-based operation will be able to write data at about 200 Mb.

Moving to Commodity

The business model that Unity is following is almost as intriguing as the technology story because both are strategically aimed at making CMOx a commodity technology that will be produced by major semiconductor manufacturers, not limited to a proprietary process totally owned by and exclusive to Unity. That involves a strategy of joint process development and IP licensing aimed at a specific vision of a mass market. That mass market turns out to be cloud computing that is accessed primarily by mobile devices, and bringing memory costs down to cents per gigabyte.

According to Eggleston, the Cloud represents a huge potential memory market but one with quite different requirements on memory capacity and performance than today. When seen as an extension of mobile devices—both consumer, such as smart phones and tablets and more specialized mobile devices—the Cloud has the potential for petabytes of fast storage. But the emphasis is on fast, as in read/write speed, and also for fast search ability. Consumers especially will see the Cloud simply as an extension to their mobile devices, which today are capable of something like 8 to 64 Gbytes of internal memory. The experience of interacting with the Cloud should feel similar to their interaction with their devices. “Consumer expectations must be met,” Eggleston says, “or they’ll reject the notion of cloud computing. NAND is not a solution and that creates an opportunity.”

Unity is therefore moving to leverage its patented technology in partnership with really big players, the first of which is Micron Technology. Having done years of basic research and development and accumulating some 120 awarded and pending patents, Unity has entered into a joint development partnership with Micron. The attraction to a commodity memory vendor is the idea that basically the same fab equipment can be used to produce a technology that will at least double if not triple the bit output per wafer of the fab facility. Under the partnership agreement, Micron is adapting the “recipe” to its fab equipment to be able to manufacture the devices. Micron has a two-year exclusive head start, has acquired an ownership stake in Unity and will receive very favorable royalty rates.

The commercialization stage will see other license agreements with large memory players such as Samsung and Toshiba. Eggleston emphasizes, “To be successful, it’s got to be a commodity. We’ve never seen a memory technology successful where only one company has it.” So CMOx is going to be going after the NAND markets not only in mobile handsets and tablets but also in mass storage systems like enterprise SSDs as well as in what seems to be the Holy Grail, Cloud computing. In addition, it is to be presumed that licensees of the scale of those being targeted will not simply remain customers but will also develop innovations based on the underlying technology. Unity already expects, once the commodity business appears to be on track, to begin developing specialty chips based on CMOx technology.

If CMOx really is successful in its targeted applications and markets, it will mean that further developments will undoubtedly be the most fascinating to watch. Also, the prediction here is that it will be subject to the Williams Law of Technology Utilization, which clearly states, “The original inventor of a new technology never has the faintest idea of how it will ultimately be used.” Stay tuned.  


Unity Semiconductor

Sunnyvale, CA.

(408) 737-7200.




Micron Technology

Boise, ID.

(208) 368-4000.