Intrinsic Semiconductor Technologies said it has successfully scaled its silicon oxide-based resistive random access memory devices (RRAM) and demonstrated electrical performance characteristics that will enable their use as high-performance, low-cost, embedded, non-volatile memory in logic devices at advanced processing nodes.
Together with its partner imec in Belgium, Intrinsic’s RRAM devices have been scaled to dimensions of 50 nm and the company said the devices have demonstrated excellent switching behavior, which is key to their use as the next generation of non-volatile, solid-state memory. The devices are compatible with the advanced semiconductor manufacturing process nodes used across the semiconductor industry, both in terms of physical dimensions (scaling) and electrical performance characteristics, making them suitable for use in edge AI and IoT applications.
Mark Dickinson, CEO of Intrinsic, said: “We are delighted to have hit this critical milestone, confirming our theoretical analysis that the devices can be made with nanoscale dimensions. For the semiconductor industry, this means, at last, there will be a simple and low-cost way to integrate non-volatile memory in any chip.”
Using Intrinsic’s technology, any chip designer will be able to embed a non-volatile memory that is as fast to read as static random access memory (SRAM), but at a fraction of the cost and power consumption. These patented “memristor” devices (a resistor with memory), which are manufactured on 300mm silicon wafers using industry standard complementary metal oxide semiconductor (CMOS) materials and processes, will enable a new era of embedded, non-volatile memory, Intrinsic claims. This has applications across the board, from enabling cheaper and more efficient microcontroller units to shifting the paradigm when designing artificial neural networks.
“It is the simplicity of our device that really makes it stand out from other emerging memories,” explained Dickinson. “We use materials and processes that are commonplace in CMOS manufacturing to create these exceptional memory devices. Our mission is to make a fast and low-cost, embedded, non-volatile memory available to any chip architect who wants to use it. In particular, applications such as edge AI or IoT, where low power consumption and high performance are key, can move from using an expensive and power-hungry external memory to a fully integrated single-chip architecture.”
Intrinsic’s device is based on a silicon oxide switching material. A conductive path between the two electrodes is formed by creating a filament of ‘oxygen vacancies’. By varying the amount of oxygen in the filament, the conductivity of the filament can be reversibly changed between a low and a high resistance state, or many states in-between.
Given the filaments are sites of oxygen vacancy rather than atoms of an introduced metal as in some memristors, the filament does not defuse into the bulk of the silicon oxide. It stays in its defined state until oxygen is forced in or removed by the set/reset process. This has two fundamental benefits. First, the filament is shown to be stable at 270 degrees C and higher, more than adequate for solder reflow temperatures. Secondly, there is no additional material to bring into the semiconductor processing and therefore no possibility of contamination, which is a real issue when working with new materials.
Nigel Toon, advisor to Intrinsic and CEO, Graphcore, said, “Intrinsic is on track to offer a new, embedded, non-volatile memory that is compatible with the most advanced semiconductor process nodes, an option that doesn’t exist today.”
Intrinsic has partnered with imec in Belgium to develop these volatile memory devices using industry standard CMOS processing. “Imec is excited to be working alongside Intrinsic to demonstrate the performance and manufacturability of their RRAM devices,” said Gouri San Kar, memory director at imec. “We are delighted to be working with such a strong team at Intrinsic, with deep knowledge of the material science behind this promising and relevant technology.”
Established in 2017, Intrinsic Semiconductor Technologies is spinout from UCL in London, U.K., the startup aims to commercialise novel memristive RRAM devices developed by Professor Tony Kenyon and Adnan Mehonic at the UCL department of electronic and electrical engineering, using their patented IP. Based in London, Intrinsic is currently developing commercial prototypes of its new memory devices, with the aim of establishing its proven designs as licensable IP for chip manufacturers.
- Resistive memories enable bio-inspired architectures for edge AI
- MRAM and ReRAM expected to displace incumbent memory technologies