Design Article
PCM Scalability:The Myth (Part 2)
Ronald Neale
10/24/2010 8:04 AM EDT
The latest PCM write/erase lifetime data
As a result of the reliability claims made for the trench structure in the form of a comment received from an EETimes reader, [3] was revisited, it was published after the first paper [1] was prepared. As a result three new orange curves have now been added to the original as shown in Figure 2.
The solid orange curve is for a trench type structure, extracted from [3]. Here it plots photolithographic node as a function of current density. The inwards movement of the sidewall, or reduction in the length of the "barrel" is directly linked to the lithographic node and can also be considered a scaled variable. To try and compare the trench structure with the "pore," the dashed orange curves in Figure 2 are for a circular contact of area equal to the rectangular heater/contact area. These curves were calculated using lithographic node to heater thickness ratios of 5 and 10.
It is not unexpected that the form of the new solid curve has more in common with those for the "link" or "gap" structures than the pore structures. Like the link device, during reset the dome PCM structure has molten chalcogenide in contact with many dielectric surfaces. With scaling, as with other types of PCM structures, the close thermal coupling will require higher reset current density to create the initiating molten hotspot.
If the pore equivalence approach has any validity, it could be claimed, based on third-party proven reliability results for array devices, that the performance is representative of pore device operation at current densities predicted by [1] for pore devices (at much lower lithographic nodes than are possible at the present time). Regardless of which one of the dashed orange curves is used, the current density for 25nm is close to 1 x10E8 Amps/sq-cm. However, it is extremely important, when attempting device-to-device comparisons, to remember that for the trench type device it is the contact current density. In the active material, the current density is decreasing as the inverse square of the radius of the dome (barrel). If the diameter of the dome is 5 times the thickness of the heater electrode, then the current density at the outer surface of the dome will decrease by a factor of about 1/25 or from 1 x10E8 A/sq-cm to 4 X 10E6 Amps/sq-cm. Therefore, any claims that levels of operating current density in the active material predicted for sub 30nm PCM devices are already reliably in use are less than completely accurate.
As a result of the reliability claims made for the trench structure in the form of a comment received from an EETimes reader, [3] was revisited, it was published after the first paper [1] was prepared. As a result three new orange curves have now been added to the original as shown in Figure 2.
The solid orange curve is for a trench type structure, extracted from [3]. Here it plots photolithographic node as a function of current density. The inwards movement of the sidewall, or reduction in the length of the "barrel" is directly linked to the lithographic node and can also be considered a scaled variable. To try and compare the trench structure with the "pore," the dashed orange curves in Figure 2 are for a circular contact of area equal to the rectangular heater/contact area. These curves were calculated using lithographic node to heater thickness ratios of 5 and 10.
It is not unexpected that the form of the new solid curve has more in common with those for the "link" or "gap" structures than the pore structures. Like the link device, during reset the dome PCM structure has molten chalcogenide in contact with many dielectric surfaces. With scaling, as with other types of PCM structures, the close thermal coupling will require higher reset current density to create the initiating molten hotspot.
If the pore equivalence approach has any validity, it could be claimed, based on third-party proven reliability results for array devices, that the performance is representative of pore device operation at current densities predicted by [1] for pore devices (at much lower lithographic nodes than are possible at the present time). Regardless of which one of the dashed orange curves is used, the current density for 25nm is close to 1 x10E8 Amps/sq-cm. However, it is extremely important, when attempting device-to-device comparisons, to remember that for the trench type device it is the contact current density. In the active material, the current density is decreasing as the inverse square of the radius of the dome (barrel). If the diameter of the dome is 5 times the thickness of the heater electrode, then the current density at the outer surface of the dome will decrease by a factor of about 1/25 or from 1 x10E8 A/sq-cm to 4 X 10E6 Amps/sq-cm. Therefore, any claims that levels of operating current density in the active material predicted for sub 30nm PCM devices are already reliably in use are less than completely accurate.
|
|
|
|
|
Navigate to related information
Volatile Memory
10/27/2010 3:22 PM EDT
Mr. Neale, an undisputed pioneer of PCM research, has done it again. In this sharp, elegant analysis in two parts, 40 years after the original article he co-authored was published, he delivers a devastating blow to the pie-in-the-sky scenarios floated around for years by unscrupulous PCM proponents.
Regarding all those techno-Ponzi schemers, who are still spreading misinformation instead of being ashamed of themselves: hoist by their own petard, indeed!
While EETimes Design should be criticized for providing a platform for Mr. Atwood and his co-conspirators, it should be highly commended for publishing Mr. Neale's analysis.
Sign in to Reply