Newest Chips Are Stacking the Decks—With Memory

RSS
Download this article in .PDF format
This file type includes high resolution graphics and schematics when applicable.

3D chip technology is pushing the envelope in the memory industry. The latest multilayer non-volatile solutions include 64-layer NAND flash memory from Toshiba, Western Digital, and Micron. Toshiba is sampling its 64-layer, BiCS Flash that uses its triple-level cell (TLC) technology. A single chip packs 32 Gbytes of storage. Chips with twice that capacity are on the roadmap using the same 64-layer architecture.

Samsung’s 128 Gbyte DDR4 DIMM

Stacking DRAM has been part of the high-bandwidth memory (HBM) systems found in high-end GPUs like AMD’s R9 Radeon and NVidia’s latest Tesla P100. HBM takes advantage of both 3D stacking and a wider, on-chip memory bus, greatly increasing bandwidth and capacity.

DRAM is not to be outdone, but it is trailing in stack size. Samsung’s latest 128-Gbyte DDR4 memory (see photo) uses a four-level stack. Many multilayer memory devices employ a bonded stack where each chip is wired to the base in the same fashion that a single die is wired. This requires that connectivity be done at the edge of the chips. Newer devices, including Samsung’s DRAM, employ through-silicon vias (TSVs). This is comparable to through-holes in a printed circuit board (PCB). It has multiple advantages, including the ability to place holes almost anywhere on the die. Samsung’s roadmap has eight- and 16-layer devices on it.

The number of layers varies with each type of device because there are other issues than just 3D stack connectivity involved in the design process. For example, DRAM operates at faster speeds than flash memory.

Still, there are more advantages driving the migration to higher stacks, including power savings and increased bandwidth. The trend toward massive in-memory databases is pushing capacities on the server side, while more compact platforms are pushing the technology in the embedded space.

There may be a limit to the number of layers that are practical, but for now we have not reached that limit.

Download this article in .PDF format
This file type includes high resolution graphics and schematics when applicable.

Please or Register to post comments.

What's alt.embedded?

Blogs focusing on embedded, software and systems

Contributors

William Wong

Bill Wong covers Digital, Embedded, Systems and Software topics at Electronic Design. He writes a number of columns, including Lab Bench and alt.embedded, plus Bill's Workbench hands-on column....
Commentaries and Blogs
Guest Blogs
Jan 26, 2017
blog

An Amateur’s View on the P2 (Part 2): Slew Rate and the Oscillator 3

Justin Mamaradlo takes a further look into the P2 op amp and how it functions, analyzing the oscillation and slew-rate characteristics of the venerable component....More
Jul 15, 2016
blog

Simple Yet Effective ESD Testing Methods for Higher Reliability 11

There are multiple ways to test for electrostatic discharge, ranging from implementing a human-body or machine model to....using a balloon and a comb?...More
Apr 8, 2016
Commentary

Confabbing on the Fabless Fad 5

High capital and maintenance costs, and EDA advances along with abstractions to deal with chip complexity, have been leading contributors to the fabless migration....More

Sponsored Introduction Continue on to (or wait seconds) ×