This slide's my new North Star, fam. IonQ's stopped thinking in silos. Compute, networking, sensing, security are not separate pillars anymore, they're runtime gears feeding each other. IonQ's wiring coherence across the stack. It's a tech company turning into infrastructure.
$IONQ
Mihir Bhaskar, IonQ's SVP of Global R&D, spoke on a defense podcast last Friday about how IonQ builds its quantum computers.
The chips inside an IonQ ion trap are built using standard CMOS manufacturing. CMOS is the same process used to make the chips in laptops, phones, and data center servers.
From Bhaskar: "There was a little bit of integration pain to get to that point. We had to take some harder routes in order to get to that point. But we did that because if you don't rely on the standard CMOS manufacturing as you scale up the effort... we just can't get the volume of chips on the timeline that are required to power our quantum computers if we were using a non-standard process."
Standard manufacturing means you can buy chips at the volumes you need when you need them. Non-standard means custom production runs that are slower and harder to scale.
He tied that choice to the SkyWater acquisition. "A large thesis of IonQ's planned acquisition of SkyWater is to be able to double down and ensure that we have a manufacturer that is supporting the quantum industry and making chips for quantum in the United States."
IonQ chose CMOS to be able to scale. SkyWater is the manufacturer.
$IONQ
IonQ was issued a patent titled Parabolic Cassegrain-Type Reflector for Ablation Loading. Co-founder Jungsang Kim is one of the inventors.
To load atoms into an ion trap, you fire a high-power laser at a piece of source material. The laser knocks atoms off the material as a cloud. The cloud drifts into the trap and the atoms get held in place.
The laser is powerful. If any of that light hits the trap itself, it heats the trap, physically damages it, or charges it. Any of those breaks the trap.
This patent places a curved mirror between the laser and the trap. The mirror has a hole in the middle. The shape is called a parabolic Cassegrain reflector, the same kind of mirror used in telescopes and satellite dishes.
The laser comes in at low intensity. The mirror focuses it down to high intensity at the source. The source is placed at the focal point and blocks any laser light from going through the hole. Only the cloud of atoms passes through. The trap never sees the high-intensity light.
This is the second IonQ patent issued on ion loading. The first made the loading process use less source material. This one keeps the trap from getting damaged during loading.
Both patents are part of making IonQ machines cheaper to build at scale. A trap that lasts longer needs to be replaced less often.
$IONQ 👀
A new paper from IonQ scientists and academic collaborators posted yesterday.
The IonQ co-authors are Martin Roetteler (VP Quantum Solutions), Panagiotis Barkoutsos, and Masako Yamada.
The team trained a quantum neural network directly on IonQ Forte Enterprise, at 16 qubits, on real patient data.
The dataset was MIMIC-III, the standard clinical machine learning benchmark, made up of electronic health records from intensive care unit admissions. The task was to predict patient survival.
The hybrid quantum-classical model achieved a mean AUC of 0.7147. The strongest classical neural baseline, Deep MICE, achieved 0.7176. The two are statistically equivalent at this scale.
Against a classical neural network of equivalent architecture, the hybrid model also showed lower variance across runs. In clinical machine learning, lower variance means more reliable predictions.
The training framework introduced in the paper reduces the cost of gradient estimation from quadratic to logarithmic in the number of qubits. At 32 qubits, the number of circuit evaluations per optimization step drops from 320 to 20. At 128 qubits, from 1,792 to 28. The advantage compounds with scale.
The paper also credits IonQ Forte Enterprise itself. All-to-all qubit connectivity and native long-range interactions allow the trainable gates to be implemented without compilation overhead.
Inference at 32 qubits is also tested on hardware. The trained models stay hardware-compatible at that scale.
IonQ trained a quantum machine learning model on patient survival data, on hardware, and matched the strongest classical baseline. The paper identifies 128 qubits as the next milestone for the task.
$IONQ
Very cool to see IonQ in the Fitz-Gerald Must Have Portfolio® ETF.
Keith pounded the table early and publicly on NVIDIA, TSLA, and PLTR.
Now IonQ is in the portfolio. 👀
$IONQ
In his shareholder letter, Niccolo de Masi gives a cost target for IonQ's fully fault-tolerant quantum computer.
"Approximately $30 million, without reliance on constrained supply chains, rare-earth materials, or helium-3, and with modest power and footprint requirements."
What he is saying is that once IonQ is producing these machines at scale, each one should cost around $30 million. And IonQ's design works around three supply constraints: limited supply chains, rare-earth materials, and helium-3.
The helium-3 part is critical.
A lot of quantum computers being built today are superconducting machines. They have to operate at temperatures close to absolute zero. Getting that cold requires helium-3, a rare form of helium.
Almost all of the world's helium-3 comes as a byproduct of nuclear weapons production. Supply is limited, and there is no easy way to make more.
IonQ builds a different kind of quantum computer. Instead of superconducting circuits, IonQ uses trapped ions, which are individual atoms held in place by electric fields. No deep cooling needed. No helium-3 needed.
De Masi's $30 million target is the cost of a machine that does not depend on anything anyone can cut off.