The team from Digital University, Kerala developed and demonstrated analog and/or digital MAC processors using ReRAM in-memory logic processing. The modules of the 8×8 1T-1R array of ReRAM designed with 130nm Skywater process were used for the implementation. We call this Kairali AI processor series that rely on the modular and scalable design of MAC processing, along with RISC-V and co-processing schemes. The entire design to development was done using open source PDK, and tools.
While any architecture of neural networks can be theoretically implemented with a MAC processor, the ReRAMs have variability that can limit its practical implementations. Two approaches to building such systems include either building networks by compensating for variability or creating architectures that leverage on the variability. We identify that reservoir computing to be an ideal candidate that best makes use of variability issues. The reservoir networks are usually untrained networks that make use of the random assignment of weights. One such network is the echo state network, that can practically make use of reservoir MAC processing followed by dense network. The concept of MAC processing presented in this article is extended across multiple projects, supported by grants from Government of Kerala, as well as through MEITY projects. Currently, we work with both open source PDKs and commercial ones, in developing various chips making use of in-memory processing with ReRAMs.
The democratisation of chip design is a primary focus in this program. Maker CASS, a broader outreach initiative supporting the Maker CHIPS program, has engaged over 1000 students through workshops on open source VLSI design. Universalization of IC Design from CASS (UNIC-CASS) has adopted these courses and is made widely available to the public.
Reference:
- Alex James (2025), “Reliable Analog In-Memory Computing with Crossbars: Memristors for Analog Neural Computing”, Foundations and Trends® in Integrated Circuits and Systems: Vol. 4: No. 1, pp 1-114. http://dx.doi.org/10.1561/3500000018
- Alex James, “Maker CHIPS: Open-Source Chips Movement for Democratization of Chip Design [CASS Chapter Highlights],” in IEEE Circuits and Systems Magazine, vol. 24, no. 4, pp. 95-100, Fourthquarter 2024, doi: 10.1109/MCAS.2024.343804.
- V. P. Dinesh, A. Radhakrishnan and A. James, “Analog Spike-Based Multihead Attention With Memristor Crossbar Arrays for Text Classification,” in IEEE Transactions on Circuits and Systems for Artificial Intelligence, vol. 2, no. 1, pp. 51-63, March 2025, doi: 10.1109/TCASAI.2024.3520011.
- V. V. Nair, E. George and A. James, “Real-Time Tumor Detection Using Electromagnetic Signals With Memristive Echo State Networks,” in IEEE Internet of Things Journal, vol. 11, no. 20, pp. 33712-33721, 15 Oct.15, 2024, doi: 10.1109/JIOT.2024.3432763.
- S. Pallathuvalappil and A. James, “Rate Coding With 3D Memristor Crossbar,” in IEEE Transactions on Circuits and Systems for Artificial Intelligence, vol. 2, no. 1, pp. 25-36, March 2025, doi: 10.1109/TCASAI.2025.3532406.
- A. Gopi, S. Davar, V. V. Nair, A. Kabeer, T. Fevens and A. James, “Memristive Implementation of Variable Pixel MRI Imaging Using Skywater 130nm PDK,” 2025 IEEE International Symposium on Circuits and Systems (ISCAS), London, United Kingdom, 2025, pp. 1-5, doi: 10.1109/ISCAS56072.2025.11044259.
Group Website: Alex James
Popular Article: IEEE CAS Magazine
Monograph: http://dx.doi.org/10.1561/3500000018
UNIC-CASS IC course: https://ml.ieee-cas.org/course/search.php?search=opensource
Keywords:
Spiking Neural Network
Memristors and memristive networks
Echo State Networks
Memristor array super-resolution
Skywater 130nm PDK