With four decades of experience dissecting digital storage and memory technologies, Tom Coughlin knows precisely where to look for emerging fractures and leaps.
Speaking at a recent ACM event, part historical survey, part contemporary analysis, but mostly a high-stakes gaze forward, he outlined the forces shaping storage’s next decade.
The main catalysts include massive data growth (394 zettabytes annually by 2028), architectural shifts toward memory-centric systems like CXL, modular chiplet designs to prep us for the end of Moore’s Law, persistent non-volatile memories poised to supplant volatile predecessors, and exotic archival options like synthetic DNA and durable optical crystals.
Coughlin, compressed his 40 years of accumulated technical wisdom into a focused narrative about where we're heading, how we’ll get there, and why it matters. All of this with an eye on the necessity of deploying storage infrastructure beyond Earth itself, for an extra futuristic punch.
Coughlin kicked off, as any storage/memory session should, with comments on the sheer scale and inevitability of data growth, framing it as an exponential shift. He pointed to IDC predictions that by 2028, data generation will balloon to an almost incomprehensible 394 zettabytes annually, driven mainly by IoT devices, sensors, and an increasing reliance on AI.
He also pressed the urgency of developing efficient, economically viable storage to manage this influx, noting that this is less about storing data and more focused on intelligently managing and accessing it, especially with AI amplifying the value of long-term retention for model training and inference.
He also talked about good old fashioned magnetic storage HDDs and tape) and their continued relevance despite perceptions that these are dying wholesale.
Even though AI has long since moved on, HDDs, Coughlin explained, remain important. The latest technical innovation, heat-assisted magnetic recording (HAMR), illustrates this ongoing vitality. HAMR uses laser-induced heat to momentarily lower the coercivity of magnetic media, allowing bits to be written more densely than previously possible, potentially reaching capacities over 100 terabytes per drive within a decade.
Tape storage, equally dismissed prematurely he says, remains relevant due to its sheer archival economics. Coughlin highlighted LTO cartridges now achieving up to 50 terabytes, with future developments possibly delivering petabyte-scale capacities. He illustrated how institutions like CERN keep enormous tape libraries, reaffirming tape’s role in data-heavy research and enterprise contexts.
But where things got truly interesting was in his exploration of solid-state storage and its transformation through NAND flash. Once constrained by planar architectures, NAND evolved dramatically through vertically stacked layers (300-layer+ 3D NAND technologies are now common).
Flash’s dominance in latency-sensitive AI and enterprise workloads, he emphasized, arises from its unique performance characteristics, creating differentiated storage strata rather than direct replacements.
He also explored some emerging storage paradigms, specifically, optical and DNA storage.
Optical technologies, he says, are currently enjoying a renaissance with their promise of archival storage with unmatched durability and potentially significant economic advantages. He pointed to Microsoft's Project Silica, which records data volumetrically within quartz crystals, offering genuinely revolutionary durability. Coughlin adds that competing optical startups like Folio Photonics and Optera are leading the way in materials science, which is key to their ability to scale.
DNA storage encodes digital information within synthetic DNA strands, offering theoretically game-changing density and longevity. He acknowledged several practical hurdles (slow read/write chemical processes) but emphasized DNA’s archival potential, especially if biotech innovations dramatically accelerated sequencing tech.
Coughlin also spent time discussing how compute hardware fundamentally interfaces with memory and storage. As he said, historically, traditional Von Neumann architectures (separation of memory and compute) consumed excessive energy moving data between components. But new memory-centric architectures,(Compute Express Link or CXL) represent strategic responses.
CXL relocates memory away from individual CPUs into shared pools accessible across multiple nodes. Thie goal is to reduce system power demands and add flexibility, though he acknowledged slight latency trade-offs (approximately one nanosecond), making it suitable for many, if not all, scenarios.
Complementing this architectural shift, the NVMe protocol, initially built for SSDs, appears increasingly likely to standardize across storage types, potentially including HDDs and magnetic tape, due to its efficient PCIe-based transport, fostering a unified, performant storage interface environment.
Coughlin also looked to the end of Moore’s Law and its implications on data, focusing first on chiplets, which uses small, specialized semiconductor dies integrated through open standards like UCIe. By combining chiplets optimized individually for logic, memory, or networking, it’s possible to build super-specialized integrated packages at lower costs and with greater flexibility. This isn’t just about compute, either. He says emerging non-volatile memory technologies (e.g., MRAM, ReRAM, PCM, and FRAM), each offer advantages over traditional volatile memories (DRAM, SRAM).
Beyond the terrestrial data explosion, he explored how localized storage and computing resources would become indispensable in space exploration contexts due to high latency in communications with Earth.
He vividly described recent lunar missions embedding data storage and memory infrastructure, projecting humanity's computing demands outward into space exploration and settlement scenarios, emphasizing the urgency of digital preservation by mentioning advanced projects that employ AI-enhanced imaging techniques, such as X-ray tomography, to digitally recover data from ancient scrolls carbonized by volcanic eruptions. I
Coughlin reinforced the narrative that storage isn’t merely a technical challenge but a cultural and civilizational imperative ensuring future access to humanity’s accumulated knowledge and discoveries.