Wednesday, March 15, 2023
Memory startup Intrinsic Semiconductor Technologies has scored funding aimed at bringing its Resistive RAM to market, which it grandly states will enable a new generation of smart devices and systems with embedded intelligence.
The UK company, which was founded in 2017 by researchers from University College London (UCL), said it has secured £7 million ($8.5 million) in a funding round led by Octopus Ventures along with existing investors IP Group and the UCL Technology Fund, plus £1 million in grants from Innovate UK, the government innovation agency.
Resistive RAM (ReRAM or RRAM) is a kind of non-volatile memory that was brought to public attention by HP over a decade ago, but which never quite lived up its hype. It was supposed to offer access speeds like DRAM, but keep its contents without power, like flash memory.
Intrinsic’s technology is based on more than a decade’s research at UCL, and serendipitously sprung from research aimed into something else entirely, as per an earlier report in The Register.
The claimed advantage of Intrinsic’s technology compared with other RRAM implementations is that it uses standard semiconductor materials, namely silicon dioxide, making it CMOS compliant and therefore more cost effective for foundries to produce using existing chip manufacturing facilities.
It also makes it easier to integrate with the kind of logic circuitry used to build processors, whereas Flash memory technology is difficult and costly to integrate in the same semiconductor chip as the processor, Intrinsic said.
This will allow data-intensive applications to overcome the current memory bottleneck imposed by having to use external flash memory components, the company added, thereby delivering higher performance with reduced energy consumption.
“We believe RRAM has the potential to become the backbone for the next generation of edge and IoT computers at a time when data hungry intelligent applications are becoming more and more prevalent,” said Intrinsic CEO Mark Dickinson.
Under this view, RRAM will enable self-contained applications and devices to process more data, perhaps enabling them to operate more complex machine learning models than would otherwise have been feasible.
“This funding will play a critical role in helping us to attract highly skilled engineers to build out the commercial potential of Intrinsic,” Dickinson said.
Intrinsic reckons its RRAM tech could “read data 10x to 100x faster and write it 1,000x faster than existing solutions”, which reminded us of something. Oh yes, Intel’s 3D XPoint memory, later marketed as Optane, was supposed to be 1,000 times faster than flash when the chip giant launched it back in 2015.
The company’s investors clearly believe that Intrinsic’s RRAM can succeed where 3D XPoint failed to meet expectations. IP Group exec Lee Thornton saod: “By solving the memory bottleneck Intrinsic has achieved a major breakthrough for the use of non-volatile memory in single-chip computers.”
However, the devil could be in the detail, according to Gartner’s vice president for semiconductors and electronics, Richard Gordon.
“If you can invent a single element memory cell that is non-volatile and random access and compatible with a standard CMOS logic process, then you might have a compelling technology,” Gordon told us.
“I’m not sure just how compatible the Intrinsic RRAM technology is with standard logic – they say it uses standard, frequently used materials and is fully CMOS compliant but the devil is in the detail here - there are probably still some non-standard materials and processes involved, would be my guess,” he added.
Gordon warned that it always difficult to scale up from the lab to manufacturing in high volume, and questioned whether integrating memory with logic was really necessary.
“Why embed a relatively small amount of memory in chip when you could stack a standalone memory die in the same package? This comes down to how much memory you need and the cost comparison between SoC (System on Chip) and SiP (System in Package). If the chip including embedded memory is small enough, then it may be the better option,” he said. ®
Copyright © 2023 CST, Inc. All Rights Reserved