Phase-change random access memory

from Wikipedia, the free encyclopedia

Phase-change Random Access Memory (abbr .: PCRAM, PRAM or PCM for short, in a special version also " Ovonics Unified Memory ", OUM, or " chalcogenide RAM ", C-RAM) is a new type of non-volatile memory in electronics ( Status 2009).

The operating principle of the memory is the change in the electrical resistance of the memory material depending on whether it is in the amorphous (high resistance / RESET state ) or in the crystalline (low resistance / SET state ) phase. The material used is a chalcogenide alloy ( chalcogen compound) - similar to the material that also takes advantage of phase changes on a CD-RW or DVD-RAM to store data. The material combinations used are z. B. from germanium , antimonyand tellurium (often alloys made from the two compounds GeTe and Sb 2 Te 3 ).

Layout and function

Schematic cross section through a PCRAM cell

The basic structure of a PCRAM memory cell is initially similar to a DRAM: it consists of a selection transistor and the resistive phase change element in which the information is stored. As in DRAM, a large number of memory cells are arranged in a matrix.

The resistive memory element consists of a metallic top electrode, a metallic bottom electrode and the phase change material in between .

The change to the unset basic state (" RESET state ") takes place by amorphizing part of the phase change material. For this purpose, the material is heated by a current pulse with a higher current strength (several hundred micro amperes ) for a shorter duration (e.g. 50 nanoseconds). After the end of the pulse, the material will cool down very quickly - so quickly that it remains in the amorphous state and does not crystallize. In order to generate this current pulse effectively, a non-linearity in the current-voltage curve of amorphous chalcogenides can be used: in itself, this state is characterized by a high resistance. However, if the applied voltage exceeds a threshold voltage, the material becomes highly conductive again ( dynamic on state ).

The change back to the crystalline set state (“SET state”) is brought about by a longer current pulse (e.g. 100 nanoseconds) with a lower current strength (several tens to a few hundred microamps). As a result, the amorphous material is heated above the crystallization temperature (see glass ) and kept at this temperature until nucleation begins and crystallization takes place.

To read out the information, a voltage is applied across the resistive element, which causes such a low current intensity that the temperature in the material does not reach the level necessary for a phase change. Depending on the state, a different current flows which is used for reading.

history

The change in electrical conductivity due to a structural change in a chalcogenide (MoS 2 ) was discovered as early as the 1920s . In the 1950s, research was carried out into the semiconducting properties of crystalline and amorphous chalcogenides. Reversible phase-changing materials were investigated with their electrical and then also optical properties in the 1960s. At that time, the construction of a non-volatile memory for electronic applications on the basis of this principle was also proposed.

Then the phase change technology was first developed further with regard to its optical application and made commercially usable: for rewritable CDs (first product in 1990, Matsushita) and DVDs . Only when in the course of these developments materials were discovered that came to interesting regions with regard to writing times and flows did the phase change RAM development also gain momentum.

At the beginning of 2006, Samsung, as the world market leader in memory chips, presented a 256- Mibit prototype and in mid-2006 a 512-Mibit prototype to replace flash chips. Series production started in September 2009.

At the Intel Developer Forum 2007 in Beijing , Intel announced its own PRAM under the code name Alderstone for the second half of 2007 , which works with so-called "Ovonian" technology - derived from the discoverer of phase change materials Stanford Ovshinsky. With 128 Mibit, the first chips that were promised have significantly smaller capacities than NOR and NAND components.

IBM announced in June 2011 that they had made stable, reliable, high-performance multi-bit PCRAM.

In 2016, IBM presented a process that enables PCM cells with 3-bit storage capacity.

Possible applications

The change in resistance between the two phases is around four orders of magnitude, which allows high sampling differences ( sensing margins ). Since the phase change is very quick and (more or less) repeatable as required , the memory principle is suitable for a random access memory , i.e. as a possible replacement for DRAM and SRAM , and in particular, since it is also non-volatile, both phases (amorphous and crystalline) are stable, also as an alternative to flash memory . Since the memory element is placed in the area of ​​the metallization levels (the so-called back-end of line , BEOL) with regard to the vertical structure and the materials and processes can largely be integrated into a CMOS production, the phase change memory is not only suitable for manufacturing of memory devices, but also for embedded memory (ger .: embedded memory ), which on the same piece of silicon is realized, as the logic circuit , which makes up the block mainly.

At the moment (2009) some semiconductor companies are developing test structures and prototypes for memory modules. Samsung started mass production in September 2009 and has been delivering the memory as a 512 MiBit die - integrated in an unspecified multichipp package for mobile devices - as PRAM since April 2010. The Micron subsidiary Numonyx has also been selling PCRAM since April 2010 under the name “Phase-Change Memory” (PCM). The chips are sold in pairs with a size of 128 Mbit each.

Stage of development

Compared to other non-volatile memories in the development stage, PCRAM shows similar expected values ​​in terms of performance, long-term durability and scalability.

The biggest problem with this memory principle is the current strength required for writing : In order to be competitive, high storage capacity and small dimensions are essential, which in PCRAM and comparable technologies requires a high packing density of the circuit elements and thus also a high degree of miniaturization. The MIS transistors used in PCRAM therefore have a channel length of less than a hundred nanometers, which means that the maximum possible current intensity drops to a few hundred micro amperes .

For this reason, work is being carried out on reducing the electrical current required for writing in various ways:

  • Phase change material and its doping : Due to the exact composition of the material and the introduction of foreign substances - e.g. B. nitrogen or tin - the electrical resistance of the material can be increased.
  • Size and shape of the bottom electrode: only material in the immediate vicinity of the bottom electrode is phase-changed; therefore the bottom electrode determines the amount of phase-changing material and thus the current required for heating. On the one hand, electrodes are published that only touch the material laterally at certain points in order to have as little material as possible to be melted. On the other hand, the phase change material is deposited in so-called sub-resolution vias (openings in the insulation layer that are so small that they can no longer be defined by the photolithographic resolution, but e.g. by reflow or etch-back techniques must be further reduced).
  • Thermal insulation of the storage element: If the material is surrounded by thick metal electrodes and thermally poorly insulating material, part of the thermal energy generated to melt the material flows away without having the desired effect.
  • Using Bipolar Transistors: Few companies are evading the problem by using this more powerful transistor technique for the select transistor. For most phase change memory applications, however, this would be of no interest due to the considerable additional costs.

Approaches to multi-level storage have already been proposed:

  • Varying the conversion volume by varying the programming pulses: different programming currents expose more or less material to the phase conversion depending on the pulse. This means that more than two resistance states that can be distinguished from one another can be recorded - although the temperature dependence of the resistance then significantly narrows the sensing margin .
  • Crystal lattice-dependent resistance: Depending on the doping of the material, it can also be present in different crystalline structures (hexagonal and face-centered cubic) - depending on the temperature in the crystallization phase. The two crystal lattices are again differentiated by their electrical resistance - but only by an order of magnitude.

Individual evidence

  1. ^ AT Waterman: The Electrical Conductivity of Molybdenite . In: Physical Review . tape 21 , no. 5 , 1923, pp. 540-549 , doi : 10.1103 / PhysRev.21.540 .
  2. ^ Stanford R. Ovshinsky: Reversible Electrical Switching Phenomena in Disordered Structures . In: Physical Review Letters . tape 21 , no. 20 , 1968, p. 1450-1453 , doi : 10.1103 / PhysRevLett.21.1450 .
  3. SAMSUNG: SAMSUNG Announces Production Start-up of its Next-generation Nonvolatile Memory PRAM. (Press release) In: SAMSUNG. September 22, 2009. Retrieved October 2, 2009 .
  4. Christof Windeck: IDF: 128 Mbit phase change memory chip still 2007. (news item) In: Heise-Online. April 18, 2007, accessed May 11, 2009 .
  5. IBM develops 'instantaneous' memory, 100 times faster than flash . engadget. June 30, 2011. Retrieved June 30, 2011.
  6. Stephen Lawson: IBM may have a cheaper DRAM alternative. IBM's new version of PCM is more dense so it costs less per byte. Computerworld, May 17, 2016, accessed May 20, 2016 .
  7. https://www.heise.de/newsticker/meldung/Auch-Samsung-liefert-Phasenuebergangsspeicherchips-nun-aus-989908.html
  8. https://www.heise.de/newsticker/meldung/Phasenuebergangsspeicherchips-jetzt-als-Serienprodukte-983361.html