As artificial intelligence (AI) continues to advance, the demands for vast amounts of data storage grow exponentially. Tech giants like Phison Electronics, Solidigm, Samsung, and Western Digital are leading the charge in developing larger and more powerful solid-state drives (SSDs) to meet the ever-increasing appetites of AI workloads.
Pushing the Limits of SSD Capacity
At the recent SC24 supercomputing conference in Atlanta, Phison Electronics grabbed the spotlight with a remarkable feat: actor Hafรพรณr Jรบlรญus Bjรถrnsson, known for his role as The Mountain in Game of Thrones, deadlifted a custom barbell weighing over a thousand pounds. The weight came from hundreds of Phison’s latest SSDs, which collectively stored an astounding 280 petabytes of data.
“Without question, this is the most data lifted by a human in history,” said Andy Higginbotham, senior director of business development at Phison Electronics. This publicity stunt highlighted the company’s recent announcement of the world’s largest SSD to date, boasting a staggering 128 terabytes of storage capacity.
Not to be outdone, Solidigm quickly followed suit with its own 123 Tb SSD, while Samsung and Western Digital also unveiled similar high-capacity drives. The race to develop larger and more powerful SSDs is fueled by the power-hungry nature of AI chips, particularly graphics processing units (GPUs) used in data centers.
Balancing Power and Performance
As Roger Corell, senior director of AI and leadership marketing at Solidigm, explains, “You can see where storage requirements are going. You look at a large language model just a couple years ago, you had a half a petabyte per rack or lower. And now there’s large language models that pair with between three and three and a half petabytes per rack. Storage efficiency to enable continued scaling of AI infrastructure is really, really important.”
Crucially, these new high-capacity SSDs can be easily swapped into existing data center racks, occupying the same space and power budget as their lower-capacity predecessors, albeit slightly taller. This seamless integration provides a straightforward way to boost storage capabilities without a complete overhaul of existing infrastructure.
Innovative Approaches to Capacity Expansion
According to Corell, there are three primary vectors for innovation when it comes to increasing SSD capacity: “One is the bits per cell, and then two is how many cells can you pack in one layer, and how many layers can you stack of these memory cells.”
Solidigm’s approach focused on packing more cells per layer, moving from a 60 Tb model to their current 122 Tb device by utilizing the smallest available NAND technology and reducing the size of non-NAND components. Phison, on the other hand, transitioned from three-bit to four-bit NAND cells while also improving along the other two vectors to achieve their groundbreaking 128 Tb capacity.
While higher bit densities can result in slower write times, Allyn Malventano, senior manager of technical marketing at Phison, notes that this trade-off is often acceptable for many applications. Maintaining the same power draw as smaller devices, however, requires a delicate balance and careful tuning of performance and power consumption.
As the demand for larger SSDs continues to soar, driven by the insatiable hunger of AI workloads, industry experts predict that petabyte-scale SSDs will become a reality within this decade. The race to develop larger and more efficient storage solutions is only just beginning.