Demand for lower-cost, higher-density DRAM has never been greater as the Internet of Things (IoT), autonomous vehicles and 5G connectivity add more devices and exponentially increase data, further straining the edge and cloud computing infrastructure. IoT devices are estimated to total 500 billion by 2030 and generate a yottabyte of data per year—that’s ten to the power of 24 bytes.



图1:DRAM sh的示意图owing an array of transistors (bottom) and tall capacitors (top).

Capacitor Aspect Ratio Scaling

The storage capacitor’s charge is proportional to the surface area of the inside and outside surfaces of the deep hole. The ratio of the hole’s height to its diameter is the aspect ratio. 2D scaling of the memory cells causes the diameter to shrink and, to maintain adequate surface area and charge, the aspect ratio must increase. To continue 2D scaling, chipmakers keep squeezing the capacitor diameter, which pushes the aspect ratio to the extreme.


As high-energy ions etch the capacitor holes, they also eat away at the hard mask. The higher the aspect ratio, the greater the risk that the hard mask is eroded before the capacitor hole is fully formed, which ruins the chip.


Figure 2: Schematic of typical challenges to scaling DRAM capacitors.

What DRAM makers really need in order to continue scaling is a denser, harder mask material that erodes at a substantially slower rate than the underlying capacitor mold. This thinner hard mask would allow deeper etches along with byproduct removal for perfectly cylindrical and straight capacitor holes with evenly matched depths.

Periphery Circuitry Scaling

A second opportunity to continue DRAM shrinking is to reduce the size of the peripheral circuitry area which consists of the logic transistors and interconnect wiring that surrounds the DRAM cell array (see Figure 3). If the memory cells scale but the periphery circuitry doesn’t, then the periphery takes up an increasing percentage of the die size. The logic is important: it helps determine the performance and power consumption of today’s high-speed DDR4 and emerging DDR5 DRAMs. Each of the metal lines that connect the logic transistors to the cell areas needs to be surrounded by an insulating dielectric material to prevent interference between electrical signals, and the thickness of this dielectric is another key factor impacting scaling. For the past 25 years, DRAM makers have used one of two silicon oxides—silane and tetraethoxysilane (TEOS)—as the dielectric material.


Today, DRAM makers are facing the same dielectric scaling challenges faced by leading-edge logic makers around 20 years ago. Beyond a certain point, thinning the dielectrics brings the metal lines so close together that capacitive coupling increases. As a result, the device suffers from increased power consumption, reduced performance, excess heat and potential reliability issues. In logic scaling, the solution was new copper low-k interconnects. In DRAM, the time for a new material that allows metal lines to be placed closer together without causing signal interference is now.

New Materials Research

These DRAM scaling materials engineering challenges—involving patterning hard masks and insulating dielectrics—have been the subject of intense research. In my next blog, I’ll discuss new materials that Applied Materials has been developing to help DRAM makers extend their product roadmaps to meet the world’s growing need for affordable, high-performance DRAMs.

קבלו עדכונים ישירות למייל!

הירשמו עכשיו

רוצים להצטרף לדיון?

ניתן ליצור איתנו קשר גם באמצעות מילויטופסזה ונחזור אליכם בהקדם.