UCL’s research has followed a similar path to that of Professor James Tour at Rice University, which is also based on the ability to switch the resistance of thin layers of silicon oxide (see UK researchers follow silicon-oxide ReRAM route and Rice University: Making memory out of silicon oxide). Professor Tour’s work has been instrumental in the formation of Weebit Nano Ltd. (see Weebit moves SiOx ReRAM on to 40nm).
Intrinsic is led by Vassilios Albanis, business manager at the UCL Business incubation service, while Professor Kenyon serves as chief scientific officer and Mehonic is chief technology officer.
Silicon dioxide is used extensively as an insulator in IC fabrication but the ability to switch resistance of very thin layers of the material with low energy and make it a conductor with a higher cycling endurance than flash memory makes it a highly promising candidate for ReRAM.
Intrinsic’s approach is based on the formation of nanometre-scale conductive filaments within the SiOx. These filaments can then be made and broken through the application of voltages of less than 1.5V. The technology can also be tuned for specific applications such as: embedded memory where it offers low power, cycling endurance and scalablity; automotive applications where its high temperature operation is valued; aerospace where it offers radiation hardness; and machine learning and neuromorphic computing.
For machine learning where matrix multiplication is used in the implementation of neural networks there is the prospect that simple cross-point geometry can provide tailored precision and accuracy at low power with a memory-centric architecture. Intrinsic’s devices also display characteristics associated with neurons that can be used in neuromorphic computing. These include synaptic plasticity, spiking behaviour, the ability to process spike-encoded data; thresholding and integration functions.
Next: From embedded memory to neuromorphic systems
Instrinsic states on its website it will provide a portfolio of technologies to address the embedded memory and automotive sector and then work with partners to develop solutions for more advanced applications, up to and including neuromorphic systems.
The company is adopting an ARM-like intellectual property licensing business model and is negotiating a seed funding round, Professor Kenyon told eeNews Europe through email correspondence.
“We have demonstrated [the technology] down to 150nm, but don’t feel that for many of the applications of interest, other than potential flash replacement, that scaling is such an important issue. More important is enhanced functionality and flexibility of a core technology platform,” Professor Kenyon said.
The company has produced a 1kbit memory array as test silicon, Professor Kenyon said. “Our initial prototypes will be for embedded applications where large arrays are not needed. We’re not looking to flash replacement.” He continued: “We aim to demonstrate high yield CMOS manufacturability – process integration in a commercial fab with standard libraries – in the next 12 months. While we have already demonstrated that our technology can be fabricated using CMOS processes, we will be working on yield determination and optimisation with a foundry partner.
When asked about what differentiates Intrinsic’s technology from that of Weebit Nano Professor Kenyon said: “Ours is standard CMOS. The operating principles and underlying mechanisms of switching are totally different in our technology, and we add the possibility of light-triggered operation, which Weebit does not. We also have a long-term vision and roadmap that takes us from near-term (embedded NVM) to medium term (hardware accelerators for ML) and long-term (neuromorphic systems). The danger of pushing for more aggressive scaling to take flash head-on, as Weebit are doing, is that in the long term, ReRAM has more to offer to other applications, and flash is, after all, very very successful.”
Professor Kenyon added that flash cannot compete in hardware acceleration, low-power embedded memory for IoT, computing at the edge or neuromorphic computing. “Better to concentrate on the applications that really need the advantages that ReRAM can offer,” he concluded.
In December 2017 the UCL team published a paper in Applied Physics Letters on light-activated switching and in February 2018 on plasticity in unipolar silicon-oxide ReRAMs (see Light-activated resistance switching in SiOx RRAM devices and Spike-timing dependent plasticity in unipolar SiOx RRAM devices).
Related links and articles: