This paper introduces Block Data Representations (BDR), a framework for exploring and evaluating a wide spectrum of narrow-precision formats for deep learning. It enables comparison of popular quantization standards, and through BDR, new formats based on shared microexponents (MX) are identified, which outperform other state-of-the-art quantization approaches, including narrow-precision floating-point and block floating-point. MX utilizes multiple levels of quantization scaling with ultra-fine scaling factors based on shared microexponents in the hardware. The effectiveness of MX is demonstrated on real-world models including large-scale generative pretraining and inferencing, and production-scale recommendation systems.

Most architectures are designed to mitigate the usually undesirable phenomenon of device wearout. We take a contrarian view and harness this phenomenon to create hardware security mechanisms that resist attacks by statistically enforcing an upper bound on hardware uses, and consequently attacks. For example, let us assume that a user may log into a smartphone a maximum of 50 times a day for 5 years, resulting in approximately 91,250 legitimate uses. If we assume at least 8-character passwords and we require login (and retrieval of the storage decryption key) to traverse hardware that wears out in 91,250 uses, then an adversary has a negligible chance of successful brute-force attack before the hardware wears out, even assuming real-world password cracking by professionals. M-way replication of our hardware and periodic re-encryption of storage can increase the daily usage bound by a factor of M.

Non-volatile memories (NVMs) have attracted significant interest recently due to their high-density, low static power, and persistence. There are, however, several challenges associated with building practical systems from NVMs, including limited write endurance and long latencies. Researchers have proposed a variety of architectural techniques which can achieve different tradeoffs between lifetime, performance and energy efficiency; however, no individual technique can satisfy requirements for all applications and different objectives. Hence, we propose Memory Cocktail Therapy (MCT), a general, learning-based framework that adaptively chooses the best techniques for the current application and objectives.

Work in quantum computer architecture has focused on communication, layout and fault tolerance, largely driven by Shor's factorization algorithm. For the first time, we study a larger range of benchmarks and find that another critical issue is the generation of code sequences for quantum rotation operations. Specifically, quantum algorithms require arbitrary rotation angles, while quantum technologies and error correction codes provide only for discrete angles and operators. A sequence of quantum machine instructions must be generated to approximate the arbitrary rotation to the required precision.