13-01-2022 | | By Robin Mitchell
Recently, researchers have demonstrated a functioning UltraRAM memory cell in silicon that could be the answer to universal memory. What is universal memory, what did the researchers demonstrate, and what does this mean for computing in the future?
Of all technologies needed to make modern computers work, it can be argued that memory is one of the most important aspects of computing. Without memory, a computer is nothing more than a calculator where the answer is immediately discarded and cannot be processed any furthermore.
Memory technologies come in a wide variety of types, with each type having its own advantages and disadvantages. For example, magnetic disks are excellent for large capacity drives with long-term data storage capabilities, but they suffer from long delays in-memory access. Another example is SRAM which is excellent for speed, but it is very costly to integrate.
As such, modern computing systems integrate different memory technologies that work together to provide an optimised solution balancing cost and performance. A computers cache, the systems fastest memory, is made from SRAM but limited in size. Actively running programs are stored in DRAM for fast access, and files are stored on hard drives for long-term storage.
Universal memory is the idea that an entire computers memory space can be put onto a single storage device that would act as a cache, program space, and file space. Such a system would benefit from major performance benefits as files and programs would be able to almost instantly share information. There would be no need for long IO processes to access data from external drives.
Currently, no universal memory device exists as finding a memory technology that is cheap, non-volatile, and quick to access is impossible. DRAM is volatile and loses its data on power-down, flash is too slow and has limited write cycles, SRAM is too expensive and volatile, and magnetic disks are incredibly slow.
Recently, researchers from the University of Lancaster and the University of Warwick have successfully demonstrated a working UltraRAM memory cell in silicon. UltraRAM is a trademarked memory technology that utilises unique properties found in compound semiconductors commonly found in optoelectronics (such as Gallium-Nitride, Gallium-Arsenide, and Silicon Carbide). Storing bits in UltraRAM is done using a floating-gate design (similar to flash memory), but a triple-barrier resonant tunnelling heterostructure is used.
Such devices have been demonstrated before in gallium, but the new announcement has seen an UltraRAM cell fabricated in silicon which marks a significant step towards unified memory. The importance of silicon over gallium is that it would allow standard semiconductor processes to integrate UltraRAM into modern processors and microcontrollers.
According to the researchers, the demonstrated memory cell can store data for at least 1000 years (thanks to a triple barrier trapping charge) while switching energy orders of magnitude lower than DRAM and flash. Writing speeds have been stated to be less than 10ms when using program/erase pulses of 2.5V, and the new memory technology has a cycle endurance up to 1000 times that of flash.
While the program speed of 10ms may seem slow, the researchers noted that their device uses a 20µm feature size, meaning that capacitive scaling into the nanometres would see switching and program speeds exceed that of DRAM.
It is hard to tell if UltraRAM would become the dominant universal memory technology for several reasons. The first is that the researchers need to prove the technology at a small scale with a reduced program time. Flash memories generally have long erase/program cycles, making them inappropriate for program memory. If UltraRAM suffers from the same challenge, it will also be unsuitable for program memory.
The second challenge is getting the industry to change. Just because a universal memory technology exists does not mean that everyone will switch over. All of the world’s computing architecture is based on the segregation of memory technologies, and introducing universal memory would create a fundamentally different computer architecture. Operating systems would need to be changed at the fundamental level to account for the new memory model, compilers would need to be updated, and new programming techniques would be required.
If UltraRAM does prove itself to be a worthy contender for use as universal memory, then the computing industry would potentially undergo one of the biggest changes since the introduction of the first microprocessor.