arXiv subjects

Brandon Reagen

Publications and source records attributed to Brandon Reagen.

At least 19 recordsLinked to original sources

arXiv2026

Academia x Industry: The Role of Fundamentals for Silicon in an AI Native Era

Agentic AI is set to become one of the most transformational technologies in generations and materially change how we approach silicon design and engineering. The impact is being felt in real time amid a rapidly changing landscape, which can make it overwhelming for both silicon practitioners and academics to adapt to the AI native silicon design era. To add structure to how we navigate this transition, we provide a joint view from academia and industry silicon practitioners of the challenges, opportunities, and considerations we expect will catalyze how the community transitions into an AI native silicon future. In particular, we reemphasize the importance of core silicon design fundamentals in academic training and why they have renewed importance in research and industry practice for AI native silicon design. It is our hope that the views provided here will offer valuable and complementary perspectives to those in academia and industry to interpret, inform, and catalyze the transition to the AI native era. We expect that many similar and overlapping views will emerge, but the precise technical details will differ across stakeholders, so it is valuable for the community to amass a diversity of viewpoints.

cs.AR↗

arXiv2026

When Proofs Meet Hardware: Comparing NTT and SumCheck in Zero-Knowledge Systems

In the ZKP community, it has long been discussed that the SumCheck protocol is asymptotically more efficient than the Number Theoretic Transform (NTT), requiring only $O(N)$ arithmetic versus $O(N \log N)$. At the same time, hardware accelerator designers propose that NTT is more hardware-friendly, benefiting from locality and data reuse, while SumCheck suffers from sequential, dependent rounds. Despite these competing intuitions, the hardware-system-level trade-offs between NTT- and SumCheck-based proving primitives remain insufficiently understood. Beyond individual accelerator design, this work presents, to our knowledge, the first hardware-system-level direct comparison of NTT- and SumCheck-based proving primitives under a unified architectural framework. We study them in the context of the ZeroCheck protocol, a common building block in zkSNARKs. We implement optimized systems for both primitives. Both are evaluated under the same level on-chip SRAM and off-chip bandwidth budgets. Our results show that there is no universal winner. Generally, SumCheck outperforms NTT for high-degree polynomials. For low-degree polynomials, performance depends on memory availability: under given SRAM budgets, NTT might deliver better performance for medium-sized workloads by exploiting data reuse. These findings, bridging cryptographic protocol design and hardware architecture, offer practical guidance for understanding the proving cost of NTT- and SumCheck-based zero-knowledge proof systems.

cs.AR↗

arXiv2026

Same Architecture, Different Capacity: Optimizer-Induced Spectral Scaling Laws

Scaling laws have made language-model performance predictable from model size, data, and compute, but they typically treat the optimizer as a fixed training detail. We show that this assumption misses a fundamental axis of representation scaling: how effectively the optimizer converts added FFN width into utilized spectral capacity. Using eigenspectra of feed-forward network representations, measured through soft and hard spectral-ranks, we find that \emph{the same Transformer architecture realizes markedly different spectral scaling laws when trained with different optimizers}. Holding architecture and width schedule fixed, AdamW exhibits weak hard-rank scaling ($\beta$=0.44) on rare-token (TAIL) representations where learning is known to be hardest, whereas Muon achieves linear scaling ($\beta$=1.02) in the same regimes, a $2.3\times$ increase in the scaling exponent. This difference is not reducible to validation loss: AdamW configurations can match low-rank Dion variants in perplexity, under extended training, while exhibiting sharply different spectral geometry, demonstrating that matched loss does not imply matched representation structure. Hard--soft rank asymmetry further reveals that optimizers differ not only in how much capacity is realized, but also in how that capacity is structured across eigenmodes. To disentangle optimizer effects from architectural ones, we compare against architectural interventions (e.g., attention rank and positional encoding), and find that optimizer-induced spectral shifts often exceed the architectural effects. These results suggest optimization as a first-class axis of representation scaling, motivating optimizer--architecture co-design.

cs.LG↗

arXiv2026

NerVE: Nonlinear Eigenspectrum Dynamics in LLM Feed-Forward Networks

We introduce NerVE, a unified eigenspectral framework for understanding how feed-forward networks (FFNs) in large language models (LLMs) organize and regulate information flow in high-dimensional latent space. Despite FFNs dominating the parameter budget, their high-dimensional dynamics remain poorly understood. NerVE addresses this gap through lightweight, memory-efficient tracking of eigenspectrum dynamics via four complementary metrics: Spectral Entropy (dispersion), Participation Ratio (effective dimensionality), Eigenvalue Early Enrichment (top-heaviness), and Jensen-Shannon divergence (distributional shifts). Our key insight is that FFN nonlinearities reinject variance across eigenmodes, fundamentally governing latent dimension utilization, and that optimizer geometry strongly modulates the extent of this variance reinjection. We validate NerVE across model scales, and diverse architectural and optimizer configurations, each uniquely shaping FFN dynamics: normalization schemes controlling variance flow; FFN weight geometries constraining latent space; positional encoding and activation functions regulating information flow; and optimizer choices redistributing effective capacity across depth. Across these settings, NerVE consistently recovers stable spectral signatures that correlate with model's generalization ability and respond predictably to design choices, generalizing beyond transformer to MLP-Mixer architectures, providing actionable insights for architectural and optimizer choices beyond trial-and-error.

cs.LG↗

arXiv2025

Network and Compiler Optimizations for Efficient Linear Algebra Kernels in Private Transformer Inference

Large language model (LLM) based services are primarily structured as client-server interactions, with clients sending queries directly to cloud providers that host LLMs. This approach currently compromises data privacy as all queries must be processed in the cloud and in the clear. Fully Homomorphic Encryption (FHE) is a solution to this data privacy issue by enabling computations directly upon encrypted queries. However, running encrypted transformer inference is challenging as programmers must map standard kernels to the constrained instruction set provided by FHE. In this work, we explore implementations of linear algebra kernels needed for transformer inference in FHE and understand how network optimization can help mitigate FHE costs while remaining performant. We leverage the Orion PyTorch to FHE framework to benchmark several linear algebra kernels in order to profile two linear transformation methods, packed row and BSGS, and find that BSGS outperforms packed row methods by up to $13.7 \times$ at transformer-level scales. We also incorporate network-level pruning strategies that reduce FHE runtimes of feed forward layers by up to $11.46\times$. Furthermore, we extend Orion to include ciphertext-ciphertext matrix-matrix products, a key component in the self-attention blocks. Finally, we perform a roofline analysis of FHE primitives and encrypted linear transformations and find that (SIMD encoded) implementations are memory-bound with primitives having roughly $0.1$ integer operations per byte of DRAM traffic. These findings illustrate the need for exploring alternative encoding schemes and models of computation within CKKS to unlock scalable private transformer inference. We conduct all experiments using the Orion framework which can be found at: https://github.com/baahl-nyu/orion.

cs.CR↗

arXiv2025

A Spatial Array for Spectrally Agile Wireless Processing

Massive MIMO is a cornerstone of next-generation wireless communication, offering significant gains in capacity, reliability, and energy efficiency. However, to meet emerging demands such as high-frequency operation, wide bandwidths, co-existence, integrated sensing, and resilience to dynamic interference, future systems must exhibit both scalability and spectral agility. These requirements place increasing pressure on the underlying processing hardware to be both efficient and reconfigurable. This paper proposes a custom-designed spatial array architecture that serves as a reconfigurable, general-purpose core optimized for a class of wireless kernels that commonly arise in diverse communications and sensing tasks. The proposed spatial array is evaluated against specialized cores for each kernel using High-Level Synthesis (HLS). Both the reconfigurable and specialized designs are synthesized in a 32 nm process to assess latency, throughput, area, and power in realistic processes. The results identify conditions under which general-purpose systolic architectures can approach the efficiency of specialized cores, thereby paving the way toward more scalable and agile systems.

eess.SP↗

arXiv2025

Spectral Scaling Laws in Language Models: How Effectively Do Feed-Forward Networks Use Their Latent Space?

As large language models (LLMs) scale, the question is not only how large they become, but how much of their capacity is effectively utilized. Existing scaling laws relate model size to loss, yet overlook how components exploit their latent space. We study feed-forward networks (FFNs) and recast width selection as a spectral utilization problem. Using a lightweight diagnostic suite -- Hard Rank (participation ratio), Soft Rank (Shannon rank), Spectral Concentration, and the composite Spectral Utilization Index (SUI) -- we quantify how many latent directions are meaningfully activated across LLaMA, GPT-2, and nGPT families. Our key finding is an asymmetric spectral scaling law: soft rank follows an almost perfect power law with FFN width, while hard rank grows only sublinearly and with high variance. This asymmetry suggests that widening FFNs mostly adds low-energy tail directions, while dominant-mode subspaces saturate early. Moreover, at larger widths, variance further collapses into a narrow subspace, leaving much of the latent space under-utilized. These results recast FFN width selection as a principled trade-off between tail capacity and dominant-mode capacity, offering concrete guidance for inference-efficient LLM design.

cs.LG↗

arXiv2025

zkPHIRE: A Programmable Accelerator for ZKPs over HIgh-degRee, Expressive Gates

Zero-Knowledge Proofs (ZKPs) have emerged as a powerful tool for secure and privacy-preserving computation. ZKPs enable one party to convince another of a statement's validity without revealing anything else. This capability has profound implications in many domains, including machine learning, blockchain, image authentication, and electronic voting. Despite their potential, ZKPs have seen limited deployment because of their exceptionally high computational overhead, which manifests primarily during proof generation. To mitigate these overheads, a (growing) body of researchers has proposed hardware accelerators and GPU implementations of both kernels and complete protocols. Prior art spans a wide variety of ZKP schemes that vary significantly in computational overhead, proof size, verifier cost, protocol setup, and trust. The latest and widely used ZKP protocols are intentionally designed to balance these trade-offs. One particular challenge in modern ZKP systems is supporting complex, high-degree gates using the SumCheck protocol. We address this challenge with a novel programmable accelerator to efficiently handle arbitrary custom gates via SumCheck. Our accelerator achieves upwards of $1000\times$ geomean speedup over CPU-based SumChecks across a range of gate types. We include this unit in zkPHIRE, a programmable, full-system accelerator that accelerates the HyperPlonk protocol. zkPHIRE achieves $1486\times$ geomean speedup over CPU and $11.87\times$ geomean speedup over the state-of-the-art at iso-area. Together, these results demonstrate compelling performance while scaling to large problem sizes (upwards of $2^{30}$ constraints) and maintaining small proof sizes ($4-5$ KB).

cs.AR↗

arXiv2025

MTU: The Multifunction Tree Unit for Accelerating Zero-Knowledge Proofs

Zero-Knowledge Proofs (ZKPs) are critical for privacy-preserving techniques and verifiable computation. Many ZKP protocols rely on key kernels such as the SumCheck protocol and Merkle Tree commitments to enable their key security properties. These kernels exhibit balanced binary tree computational patterns, which enable efficient hardware acceleration. Although prior work has investigated accelerating these kernels as part of an overarching ZKP protocol, exploiting this common tree pattern remains relatively underexplored. We conduct a systematic evaluation of these tree-based workloads under different traversal strategies, analyzing performance on multi-threaded CPUs and the Multifunction Tree Unit (MTU) hardware accelerator. We introduce a hardware-friendly Hybrid Traversal for binary tree that improves parallelism and scalability while significantly reducing memory traffic on hardware. Our results show that MTU achieves up to $1478\times$ speedup over CPU at DDR-level bandwidth and that our hybrid traversal outperforms breadth-first search by up to $3\times$. These findings offer practical guidance for designing efficient hardware accelerators for ZKP workloads with binary tree structures.

cs.AR↗

arXiv2025

A Random Matrix Theory Perspective on the Learning Dynamics of Multi-head Latent Attention

In this work, we study how multi-head latent attention (MLA), a popular strategy for compressing key/value memory, affects a transformer's internal capacity during pretraining. Using a lightweight suite of Marchenko-Pastur (MP) diagnostics, we analyze the spectrum of the $W_{Q}W_{K}^\top$ gram matrix throughout training, comparing three variants: the standard multi-head attention (MHA) baseline, MLA-PreRoPE with rotary applied before compression, and MLA-Decoupled, which shares a single rotary sub-vector across all heads. Our random matrix analysis reveals \textbf{three key findings:} \textbf{ i)} capacity bottlenecks emerge locally: both MHA and MLA-PreRoPE exhibit sharp, early spikes in specific layers that persist and propagate, disrupting the balance between bulk and outlier directions; \textbf{ ii)} these spikes coincide with rank collapse, concentrating the model's expressivity into narrow subspaces; \textbf{ iii)} only the decoupled variant prevents this cascade, maintaining broad spectral support and suppressing outlier formation across layers. These results underscore that \emph{how} rotary embeddings are applied is just as critical as \emph{where} compression occurs. Sharing rotary components across heads mitigates spectral fragmentation and preserves representational capacity.

cs.LG↗

arXiv2025

EinHops: Einsum Notation for Expressive Homomorphic Operations on RNS-CKKS Tensors

Fully Homomorphic Encryption (FHE) is an encryption scheme that allows for computation to be performed directly on encrypted data, effectively closing the loop on secure and outsourced computing. Data is encrypted not only during rest and transit, but also during processing. However, FHE provides a limited instruction set: SIMD addition, SIMD multiplication, and cyclic rotation of 1-D vectors. This restriction makes performing multi-dimensional tensor operations challenging. Practitioners must pack these tensors into 1-D vectors and map tensor operations onto this one-dimensional layout rather than their traditional nested structure. And while prior systems have made significant strides in automating this process, they often hide critical packing decisions behind layers of abstraction, making debugging, optimizing, and building on top of these systems difficult. In this work, we approach multi-dimensional tensor operations in FHE through Einstein summation (einsum) notation. Einsum notation explicitly encodes dimensional structure and operations in its syntax, naturally exposing how tensors should be packed and transformed. We decompose einsum expressions into a fixed set of FHE-friendly operations. We implement our design and present EinHops, a minimalist system that factors einsum expressions into a fixed sequence of FHE operations. EinHops enables developers to perform encrypted tensor operations using FHE while maintaining full visibility into the underlying packing strategy. We evaluate EinHops on a range of tensor operations from a simple transpose to complex multi-dimensional contractions. We show that the explicit nature of einsum notation allows us to build an FHE tensor system that is simple, general, and interpretable. We open-source EinHops at the following repository: https://github.com/baahl-nyu/einhops.

cs.CR↗

arXiv2025

HE-LRM: Encrypted Deep Learning Recommendation Models using Fully Homomorphic Encryption

Fully Homomorphic Encryption (FHE) enables computation directly on encrypted data and privacy-preserving neural inference in the cloud. Existing solutions focus on models with dense inputs (e.g., CNNs and MLPs). Recommendation models (e.g., DLRM) pose a different challenge: sparse categorical inputs require private lookups into large embedding tables, which must be implemented using FHE's restrictive operators. Naive lookups incur significant communication and memory costs; prior work proposes compressing embedding tables at the expense of introducing large server-side compute costs (i.e., indicator function) and revealing embedding-table structure. We present HE-LRM, a performance optimized solution for executing recommendation with FHE. First, we develop an embedding compression technique using client-side digit decomposition that achieves 56$\times$ speedup over the state-of-the-art. Next, we propose a multi-embedding packing strategy that enables ciphertext SIMD-parallel lookups across multiple tables. We integrate HE-LRM into the open-source Orion FHE framework to demonstrate end-to-end encrypted DLRM inference. We evaluate HE-LRM on UCI (health prediction) and Criteo (click prediction), achieving inference latencies of 24 seconds on UCI and 228 to 489 seconds, respectively, on a single-threaded CPU. Finally, we show how GPU and ASIC FHE acceleration can reduce end-to-end latencies to seconds and even sub-seconds. Our code can be found at https://github.com/baahl-nyu/orion/tree/criteo-helrm.

cs.CR↗

arXiv2025

Need for zkSpeed: Accelerating HyperPlonk for Zero-Knowledge Proofs

Zero-Knowledge Proofs (ZKPs) are rapidly gaining importance in privacy-preserving and verifiable computing. ZKPs enable a proving party to prove the truth of a statement to a verifying party without revealing anything else. ZKPs have applications in blockchain technologies, verifiable machine learning, and electronic voting, but have yet to see widespread adoption due to the computational complexity of the proving process. Recent works have accelerated the key primitives of state-of-the-art ZKP protocols on GPU and ASIC. However, the protocols accelerated thus far face one of two challenges: they either require a trusted setup for each application, or they generate larger proof sizes with higher verification costs, limiting their applicability in scenarios with numerous verifiers or strict verification time constraints. This work presents an accelerator, zkSpeed, for HyperPlonk, a state-of-the-art ZKP protocol that supports both one-time, universal setup and small proof sizes for typical ZKP applications in publicly verifiable, consensus-based systems. We accelerate the entire protocol, including two major primitives: SumCheck and Multi-scalar Multiplications (MSMs). We develop a full-chip architecture using 366.46 mm$^2$ and 2 TB/s of bandwidth to accelerate the entire proof generation process, achieving geometric mean speedups of 801$\times$ over CPU baselines.

cs.AR↗

arXiv2025

Entropy-Guided Attention for Private LLMs

The pervasiveness of proprietary language models has raised critical privacy concerns, necessitating advancements in private inference (PI), where computations are performed directly on encrypted data without revealing users' sensitive information. While PI offers a promising solution, its practical deployment is hindered by substantial communication and latency overheads, primarily stemming from nonlinear operations. To address this, we introduce an information-theoretic framework to characterize the role of nonlinearities in decoder-only language models, laying a principled foundation for optimizing transformer-architectures tailored to the demands of PI. By leveraging Shannon's entropy as a quantitative measure, we uncover the previously unexplored dual significance of nonlinearities: beyond ensuring training stability, they are crucial for maintaining attention head diversity. Specifically, we find that their removal triggers two critical failure modes: {\em entropy collapse} in deeper layers that destabilizes training, and {\em entropic overload} in earlier layers that leads to under-utilization of Multi-Head Attention's (MHA) representational capacity. We propose an entropy-guided attention mechanism paired with a novel entropy regularization technique to mitigate entropic overload. Additionally, we explore PI-friendly alternatives to layer normalization for preventing entropy collapse and stabilizing the training of LLMs with reduced-nonlinearities. Our study bridges the gap between information theory and architectural design, establishing entropy dynamics as a principled guide for developing efficient PI architectures. The code and implementation are available at https://github.com/Nandan91/entropy-guided-attention-llm

cs.LG↗

arXiv2024

TruncFormer: Private LLM Inference Using Only Truncations

Private inference (PI) serves an important role in guaranteeing the privacy of user data when interfacing with proprietary machine learning models such as LLMs. However, PI remains practically intractable due to the massive latency costs associated with nonlinear functions present in LLMs. Existing works have focused on improving latency of specific LLM nonlinearities (such as the Softmax, or the GeLU) via approximations. However, new types of nonlinearities are regularly introduced with new LLM architectures, and this has led to a constant game of catch-up where PI researchers attempt to optimize the newest nonlinear function. We introduce TruncFormer, a framework for taking any LLM and transforming it into a plaintext emulation of PI. Our framework leverages the fact that nonlinearities in LLMs are differentiable and can be accurately approximated with a sequence of additions, multiplications, and truncations. Further, we decouple the add/multiply and truncation operations, and statically determine where truncations should be inserted based on a given field size and input representation size. This leads to latency improvements over existing cryptographic protocols that enforce truncation after every multiplication operation. We open source our code for community use.

cs.CR↗

arXiv2024

AERO: Entropy-Guided Framework for Private LLM Inference

Privacy-preserving computation enables language model inference directly on encrypted data yet suffers from prohibitive latency and communication overheads, primarily due to nonlinear functions. Removing nonlinearities, however, can trigger one of two failure modes restricting the potential for nonlinearity removal: entropy collapse in deeper layers, which destabilizes training, and entropic overload in early layers, causing under-utilization of attention heads. To address these challenges, we introduce AERO, an entropy-guided framework to strategically eliminates costly nonlinear operations from transformer architectures, which employs an adaptive recalibration through a head-wise entropy regularizer with learnable per-head strengths, enabling each head to adjust its entropy level while penalizing extreme entropies and fostering functional diversity through a tolerance margin. Experiments show AERO can save 3.4$\times$ communication and 1.4$\times$ latency, without any performance penalty.

cs.LG↗

arXiv2024

ReLU's Revival: On the Entropic Overload in Normalization-Free Large Language Models

LayerNorm is a critical component in modern large language models (LLMs) for stabilizing training and ensuring smooth optimization. However, it introduces significant challenges in mechanistic interpretability, outlier feature suppression, faithful signal propagation, and computational and communication complexity of private inference. This work explores desirable activation functions in normalization-free decoder-only LLMs. Contrary to the conventional preference for the GELU in transformer-based models, our empirical findings demonstrate an {\em opposite trend} -- ReLU significantly outperforms GELU in LayerNorm-free models, leading to an {\bf 8.2\%} perplexity improvement. We discover a key issue with GELU, where early layers experience entropic overload, leading to the under-utilization of the representational capacity of attention heads. This highlights that smoother activations like GELU are {\em ill-suited} for LayerNorm-free architectures, whereas ReLU's geometrical properties -- specialization in input space and intra-class selectivity -- lead to improved learning dynamics and better information retention in the absence of LayerNorm. This study offers key insights for optimizing transformer architectures where LayerNorm introduces significant challenges. The code and implementation are available at https://github.com/Nandan91/relu-revival-normfree

cs.LG↗

arXiv2024

Osiris: A Systolic Approach to Accelerating Fully Homomorphic Encryption

In this paper we show how fully homomorphic encryption (FHE) can be accelerated using a systolic architecture. We begin by analyzing FHE algorithms and then develop systolic or systolic-esque units for each major kernel. Connecting units is challenging due to the different data access and computational patterns of the kernels. We overcome this by proposing a new data tiling technique that we name limb interleaving. Limb interleaving creates a common data input/output pattern across all kernels that allows the entire architecture, named Osiris, to operate in lockstep. Osiris is capable of processing key-switches, bootstrapping, and full neural network inferences with high utilization across a range of FHE parameters. To achieve high performance, we propose a new giant-step centric (GSC) dataflow that efficiently maps state-of-the-art FHE matrix-vector product algorithms onto Osiris by optimizing for reuse and parallelism. Our evaluation of Osiris shows it outperforms the prior state-of-the-art accelerator on all standard benchmarks.

cs.CR↗