The SpinAge consortium is advancing the monolithic integration of CMOS, spintronics, and photonics to enable neuromorphic computing platforms based on spin torque and spin Hall nano-oscillators (STNOs/SHNOs).
A major focus has been the design and implementation of CMOS circuits together with spintronic and photonic layers. Particular effort went into addressing alignment challenges between laser and spintronic layers, where microlenses were developed to ensure robust focusing of the laser beam. Interconnection and routing between the three layers were also carefully engineered, including the use of specific CMOS metal layers for spintronic device fabrication.
At the wafer level, CMOS and VCSEL designs were merged with MTJ-based spintronic devices into a unified reticle and submitted for fabrication. While delivery from the foundry is pending, emulation and Cadence simulations have already demonstrated system viability, replicating both spintronic dynamics and laser effects.
On the computing side, spintronic nano-oscillators were validated as fundamental building blocks of neuromorphic architectures. Experimental work showed that MRAM–STNO systems can achieve programmable multi-state behavior, scaling from minimal 2-input configurations up to 32 programmable resistance states with larger MTJ series. Neuromorphic simulations confirmed that oscillator-based networks can achieve software-comparable accuracy (>95%) on benchmark datasets such as IRIS.
Further advances combined optical and electrical driving of MTJs, demonstrating that laser heating enhances nonlinear dynamics and enables functionalities such as multi-level states and voltage spiking, effectively mimicking neural behavior. Material innovations, such as Bi-doped Pt alloys, were also explored to improve SHNO efficiency, though with tradeoffs in output power and linewidth.
For signal readout and system integration, new RF front-ends were developed, including active baluns, multi-stage LNAs, and nonlinear detection circuits, improving the ability to capture weak SHNO signals. Progress was also made on in-memory computing with memristor crossbars, using both stacked-die and flip-chip assembly approaches. Although packaging challenges delayed complete demonstrations, prior CMOS+RRAM platforms were successfully used for pattern recognition experiments.
The project also achieved advances in cross-layer integration: development of microlenses and VCSEL flip-chip arrays, enhancement of Seebeck effects in STNOs with transparent contacts under bias, and upgrades to magneto-dynamical measurement systems for evaluating integrated devices.
Overall Outcome
The work demonstrates strong progress in both device-level validation (STNO/SHNO, MRAM, RRAM, VCSELs) and system-level integration (monolithic CMOS–spintronics–photonics platform). While some hardware demonstrations await wafer delivery and packaging solutions, simulations, partial prototypes, and joint integration efforts confirm the feasibility of SpinAge’s neuromorphic computing concepts. The final demonstration (monolithically integrated) is ongoing and it will be finalized very soon (CMOS wafer is ready, Photonic VCSEL array is ready and the Spintronic is ready).