A Chip That Works at 700 Degrees Celsius Has Been Developed in California 0 11

Typically, electronics in a phone, car, or space satellite function until their temperature becomes critically high. Around 200 degrees Celsius, conventional silicon circuits stop working. This threshold has been considered nearly insurmountable for decades. However, engineers from the University of Southern California seem to have found a way to bypass this limitation.

They have created a memristor that continues to operate reliably at 700 degrees, which is hotter than the temperature of molten lava. And, according to the data, this is not the limit, but simply the ceiling of their measuring setup.

The research was published in the journal Science under the guidance of Professor Joshua Yang. The device is structured like a nanoscale sandwich. The top layer contains tungsten, a metal with the highest melting point, the middle layer is made of hafnium oxide ceramic, and the bottom layer consists of a graphene layer that is one atom thick. In this configuration, the memory retains data for over fifty hours at 700 degrees without refreshing, withstands more than a billion rewrite cycles, and operates at a voltage of just 1.5 volts with speeds in the tens of nanoseconds.

Interestingly, this result was achieved accidentally. Initially, Yang's group was trying to implement a different idea using graphene. That did not work, but something unexpected was discovered. Conventional memristors based on platinum and hafnium oxide fail when heated because tungsten atoms begin to migrate through the ceramic layer until they reach the bottom electrode and short-circuit the device. In the new design, graphene halts this migration. The surface interaction between graphene and tungsten turned out to be so weak that tungsten atoms have nothing to latch onto. They simply cannot form a continuous conductive bridge, meaning that a short circuit does not occur.

To confirm the mechanism, the team used transmission electron microscopy, spectroscopy, and quantum modeling. It turned out that at the interface between graphene and tungsten, the atoms behave differently than on conventional metallic electrodes. The barriers for surface diffusion are higher there, and adsorption is weaker.

Now, knowing this principle, it is possible to search for other materials with similar surface chemistry, which in the long run will simplify industrial production.

There are several potential applications for such heat-resistant memory. Space agencies have long dreamed of electronics capable of operating at the surface temperature of Venus, where even the most protected probes fail. Geothermal energy requires sensors that can descend into deep wells, where extreme temperatures prevail. Nuclear and thermonuclear facilities also generate significant heat for their control equipment. And for automotive electronics, designed for peak temperatures of 125 degrees, a chip that remains operational at 700 would be virtually indestructible.

The device is of particular interest in the context of artificial intelligence. Memristors can perform the main operation of neural networks, matrix multiplication, not like conventional digital processors, which iterate through numbers step by step, but physically, according to Ohm's law. Voltage multiplied by conductivity gives current. And this current is already the result of the computation. Joshua Yang explains that over 92% of the computational load of systems like ChatGPT is related to matrix multiplication. This means that such chips could perform it orders of magnitude faster and with much lower energy consumption.

Three co-authors of the paper, along with Yang, have already founded a startup called TetraMem, which is commercializing memristor chips for AI at room temperature. The lab already has working chips that students use for machine learning tasks. A high-temperature version could expand this technology into environments where conventional electronics do not survive.

Of course, there is still a long way to go before a finished product is available. Memory alone does not make a computer complete – high-temperature logic circuits are also needed, and current devices are manually assembled in the lab at a submicron scale. The market introduction of the technology will take time, but the tungsten and hafnium oxide used have long been employed in the semiconductor industry, and although graphene is relatively new to the field, TSMC and Samsung have included it in their roadmaps and have already learned to grow it on silicon wafers.

Yang himself calls this the first step. According to him, the missing component has now been created, meaning that the possibility exists. It remains only to turn it into real devices that can operate where electronics previously just melted.