Design Article
PCM Progress: Temperatures rise and constituents on the move
R.G. Neale
1/20/2011 11:26 AM EST
For those, both optimists and pessimists, who follow phase change memory (PCM) developments, the close of the year and IEDM2010 offered something for both camps. Two papers in particular [1], [2], provided the not all good news, not all bad news contrasts, and here we will explore some of the significant points of each. Along the way, I will review the impact both might have on the overall picture of PCM progress.
Raising the PCM data retention temperature
A quick overview of [1] and its positive conclusions, has one wondering why we have not had announcements from the companies attempting to commercialize PCM devices that they will make an immediate switch to GeTe(2% nitrogen). If the move has already been made, the results have not yet appeared on any published data sheets.
One of the recognized challenges for PCM is the problem of data retention at elevated temperatures, caused by the crystallization of amorphous state, resulting in data loss. The most used germanium, antimony, tellurium (GST) compositions are claimed in "commercial" products to be capable of operation at 85oC for ten years. The associated failure rates in ppb, as a function of write/erase lifetime do not appear to be readily available.
For those with concerns about composition changes caused by PCM operation at current densities of the order 2x 10E7A/sq-cm and higher, it is more than just the starting composition that must be resistant to operation at elevated temperature. The change in composition with write/erase (w/e) lifetime, or accumulated number of w/e cycles must also be considered, as will be shown here in later paragraphs.
Ref [1] now reports that a move away from GST and pure GeTe to PCM materials based on the nitrogen "doping" of GeTe from 2% to10% appears to provide PCM materials that have significantly improved performance with respect to memory data retention.
GeTeN PCM device testing
The important and very significant improvement reported by [1], is that the crystallization temperature of the GeTeNx% materials is increased from 180oC for pure GeTe up to 269oC for a 10% nitrogen doping level, with a melting temperature of 720oC for all levels of doping . For comparison, the melting temperature of GST is about 610oC, supported by claims of superior write/erase time characteristics.
The testing of the doped materials at device level was based on "lance" structures (see Figure 1) with a minimum contact diameter of 300nm. These devices required a reset current of between 20 and 30mA of 100ns duration. Most of the experimental device work used a reset pulse of 30mA.
The optimum material appears to be GeTeN2%. For this material, as well as characterization, the key device-based experiment was data retention at elevated temperature. In this experiment, five sets of 40 devices were operated at temperatures of from 200oC to 250oC and the change in resistance was monitored as a function of time at that temperature. This temperature range crosses the reported crystallization temperature for blanket deposited material of the same composition. For the higher temperatures, crystallization is very rapid ~100secs, and the failure criteria used was a reduction in resistance of 50%. From this experiment and similar, plots of time to failure as a function of 1/kT provided the activation energy of the crystallization process. For GeTeN2%, the reported value is 3.1eV. Extrapolation of this to 10 years leads the authors [1] to claim a new highest data retention temperature for PCM materials of 154o C.
The device work is supported by some very detailed composition structural analysis. Intuitively, and wrongly, as the experimental evidence demonstrates, it might be expected that the material with the highest crystallization temperature might be the best for elevated data retention in PCM devices.
The N-doped GeTe material with the lowest crystallization temperature (GeTeN2%) has the best data retention characteristics. In all of the N-doped family of materials, the detailed structural analysis presented in [1] very strongly suggests that amorphous GeN plays an important role in limiting crystal growth. In simple terms, it does this by collecting at grain boundaries and inhibits or slows the growth of the crystallites.
From bonding analysis, it appears amorphous GexNy, close to Ge3N4 is present in amorphous and crystallized materials, at all doping levels. Ge3N4 crystallizes at 800oC; therefore, once formed, it is likely to be present at all times. If the authors of [1] are correct in the interpretation of the peak in PCM thermal performance at the composition GeTeN2%, it is because crystallization is growth limited and nucleation inhibited. After that composition with increasing levels of nitrogen doping, multiple nucleation points determine crystal volume fraction and growth at any time.
What is missing and we hope part of the work in-progress is a determination of the shape of the distribution curve that records percentage of device failures as a function of time for a given stress (temperature) test. Also, the evaluation of other PCM characteristics in scaled devices, especially with write/erase lifetime, for reasons discussed next.
Raising the PCM data retention temperature
A quick overview of [1] and its positive conclusions, has one wondering why we have not had announcements from the companies attempting to commercialize PCM devices that they will make an immediate switch to GeTe(2% nitrogen). If the move has already been made, the results have not yet appeared on any published data sheets.
One of the recognized challenges for PCM is the problem of data retention at elevated temperatures, caused by the crystallization of amorphous state, resulting in data loss. The most used germanium, antimony, tellurium (GST) compositions are claimed in "commercial" products to be capable of operation at 85oC for ten years. The associated failure rates in ppb, as a function of write/erase lifetime do not appear to be readily available.
For those with concerns about composition changes caused by PCM operation at current densities of the order 2x 10E7A/sq-cm and higher, it is more than just the starting composition that must be resistant to operation at elevated temperature. The change in composition with write/erase (w/e) lifetime, or accumulated number of w/e cycles must also be considered, as will be shown here in later paragraphs.
Ref [1] now reports that a move away from GST and pure GeTe to PCM materials based on the nitrogen "doping" of GeTe from 2% to10% appears to provide PCM materials that have significantly improved performance with respect to memory data retention.
GeTeN PCM device testing
The important and very significant improvement reported by [1], is that the crystallization temperature of the GeTeNx% materials is increased from 180oC for pure GeTe up to 269oC for a 10% nitrogen doping level, with a melting temperature of 720oC for all levels of doping . For comparison, the melting temperature of GST is about 610oC, supported by claims of superior write/erase time characteristics.
The testing of the doped materials at device level was based on "lance" structures (see Figure 1) with a minimum contact diameter of 300nm. These devices required a reset current of between 20 and 30mA of 100ns duration. Most of the experimental device work used a reset pulse of 30mA.
The optimum material appears to be GeTeN2%. For this material, as well as characterization, the key device-based experiment was data retention at elevated temperature. In this experiment, five sets of 40 devices were operated at temperatures of from 200oC to 250oC and the change in resistance was monitored as a function of time at that temperature. This temperature range crosses the reported crystallization temperature for blanket deposited material of the same composition. For the higher temperatures, crystallization is very rapid ~100secs, and the failure criteria used was a reduction in resistance of 50%. From this experiment and similar, plots of time to failure as a function of 1/kT provided the activation energy of the crystallization process. For GeTeN2%, the reported value is 3.1eV. Extrapolation of this to 10 years leads the authors [1] to claim a new highest data retention temperature for PCM materials of 154o C.
The device work is supported by some very detailed composition structural analysis. Intuitively, and wrongly, as the experimental evidence demonstrates, it might be expected that the material with the highest crystallization temperature might be the best for elevated data retention in PCM devices.
The N-doped GeTe material with the lowest crystallization temperature (GeTeN2%) has the best data retention characteristics. In all of the N-doped family of materials, the detailed structural analysis presented in [1] very strongly suggests that amorphous GeN plays an important role in limiting crystal growth. In simple terms, it does this by collecting at grain boundaries and inhibits or slows the growth of the crystallites.
From bonding analysis, it appears amorphous GexNy, close to Ge3N4 is present in amorphous and crystallized materials, at all doping levels. Ge3N4 crystallizes at 800oC; therefore, once formed, it is likely to be present at all times. If the authors of [1] are correct in the interpretation of the peak in PCM thermal performance at the composition GeTeN2%, it is because crystallization is growth limited and nucleation inhibited. After that composition with increasing levels of nitrogen doping, multiple nucleation points determine crystal volume fraction and growth at any time.
What is missing and we hope part of the work in-progress is a determination of the shape of the distribution curve that records percentage of device failures as a function of time for a given stress (temperature) test. Also, the evaluation of other PCM characteristics in scaled devices, especially with write/erase lifetime, for reasons discussed next.
|
|
|
|
|
Navigate to related information
janine.love
1/24/2011 9:33 AM EST
The subject/approach of this article was actually suggested by a reader of Ron Neale's last PCM article for us. My thanks to Ron for taking it on and to our reader for his suggestion.
Sign in to Reply
Volatile Memory
2/23/2011 9:28 PM EST
Mr. Neale's series of four article published by EETimes over the past months will surely become a classic. The "optimists" knew the days of the Techno-Ponzi called Phase-Change Memory were numbered. Now we, the readers, know it as well.
Sign in to Reply