Jekyll2023-12-11T15:39:14+00:00https://cphenicie.github.io/feed.xmlChristopher PhenicieWrite an awesome description for your new site here. You can edit this line in _config.yml. It will appear in your document head meta (for Google search results) and in your feed.xml site description.Christopher PhenicieList of AI photonics companies2020-12-26T00:00:00+00:002020-12-26T00:00:00+00:00https://cphenicie.github.io/list-of-AI-photonics-companies<script type="text/x-mathjax-config">
MathJax.Hub.Config({
tex2jax: {
skipTags: ['script', 'noscript', 'style', 'textarea', 'pre'],
inlineMath: [['$','$']]
}
});
</script>
<script src="https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML" type="text/javascript"></script>
<p>Final update: I won’t be updating this page anymore, it’s likely very out of date. Sorry about that!</p>
<p>Last content updated Dec 26, 2020. N = 8 (+4) companies.</p>
<p>I have spent some time looking in to companies that are trying to build computer chips that using photonics. That is, chips that use at least some light in addition to electricity. Nominally these companies pitch their chips for AI uses, though I think many of these companies are pursuing chips that would be useful more general computation, as well. Note that I am not including companies making optical transceivers. (These are devices the don’t do general computation, but convert between electric signals and optical signals. This is how, for instance, internet data can be transmitted on fiber optic cables but interface with computers, which use electric signal.) I don’t know much about this area, so I will just leave the companies in alphabetical order, but I’ll separate startups from established companies. I got the names of several of these companies from <a href="https://twitter.com/jwangARK/status/1268230473421127683">James Wang’s twitter account</a></p>
<h1 id="established-companies">Established Companies</h1>
<ul>
<li><a href="https://www.labs.hpe.com/next-next/light">HP Enterprise</a></li>
<li><a href="https://www.intel.com/content/www/us/en/architecture-and-technology/silicon-photonics/silicon-photonics-overview.html">Intel</a></li>
</ul>
<h1 id="startups">Startups</h1>
<ul>
<li><a href="https://www.fathomcomputing.com/">Fathom Computing</a></li>
<li><a href="https://www.lightelligence.ai/">Lightelligence</a></li>
<li><a href="https://lightmatter.co/">Lightmatter</a></li>
<li><a href="https://lighton.ai/">LightOn</a></li>
<li><a href="https://luminous.co/">Luminous Computing</a></li>
<li><a href="https://optalysys.com/">Optalysys</a></li>
</ul>
<h1 id="quantum-computing-companies">Quantum Computing Companies</h1>
<p>There are also four companies that I know of that are pursuing photonics as a route for quantum computing, that is in turn advertised as an AI chip. I’m separating these out from the other photonics companies because they are trying to do fundamentally different computations, and are not in direct competition with the companies above. These companies are:</p>
<ul>
<li><a href="https://anametric.com/">Anametric</a></li>
<li><a href="https://psiquantum.com/">PsiQuantum</a></li>
<li><a href="https://www.quix.nl/">Quix</a></li>
<li><a href="https://www.xanadu.ai/">Xanadu</a></li>
</ul>Christopher PhenicieGenerating priors from historical data2020-12-24T00:00:00+00:002020-12-24T00:00:00+00:00https://cphenicie.github.io/generating-priors-from-historical-data<script type="text/x-mathjax-config">
MathJax.Hub.Config({
tex2jax: {
skipTags: ['script', 'noscript', 'style', 'textarea', 'pre'],
inlineMath: [['$','$']]
}
});
</script>
<script src="https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML" type="text/javascript"></script>
<p><em>Note, you can find the original Jupyter notebooks used to generate the plots in this post in my <a href="https://github.com/cphenicie/forecasts">forecasts repo</a> in the Foretell folder. The file is 20201118_value_of_Chinese_semiconductor-imports.ipynb, and there is a PDF printout of the file for ease of viewing</em></p>
<p>I have recently started forecasting on CSET’s <a href="https://www.cset-foretell.com/">Foretell</a> platform. I’ve been interested in forecasting for a while, but I’ve always gotten stuck on where to start. The advice that most people seem to give (including Foretell’s own <a href="https://www.cset-foretell.com/courses/2/slides/1">tutorial</a>) is to start with a base rate (aka an “outside view”) and then update it using the specifics of the situation you are forecasting (aka an “inside view”). The problem I’ve had with this is that I generally have neither and inside view nor an outside view for most forecasts. I generally just come in with a uniform prior, thinking that all the options of the forecast are equally likely. In this post I’ll explain my first step to go beyond this “uniform prior” forecast.</p>
<p>One nice thing about Foretell’s questions is sometimes they provide data from which one can construct a base rate. One example of this is the question <a href="https://www.cset-foretell.com/questions/95-what-will-be-the-value-in-dollars-of-all-chinese-imports-of-semiconductor-manufacturing-equipment-in-2021">What will be the value, in dollars, of all Chinese imports of semiconductor manufacturing equipment in 2021?</a> The “answer” to this question takes the form of your confidence that this dollar value will fall within the ranges (in units of billion USD) {[0, 20], (20, 30], (30, 40], (40, 50], (50, $\infty$)}</p>
<p>If you click “Show background information” below the question, they provide nine years’ worth of data measuring exactly this quantity.</p>
<p><img src="/assets/images/2020-12-24-forecast/original_data.png" alt="original data" /></p>
<p>Now, with this data in hand, I actually feel like I have a reason to start assigning numbers to an outside view. Specifically, I can make an outside view prediction that the future will just follow the trajectory of the past. The physicist in me thinks that a good first guess is we can take a linear approximation of the trend by fitting a line to it, using the functional form</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="k">def</span> <span class="nf">outside_func</span><span class="p">(</span><span class="n">t</span><span class="p">,</span> <span class="n">m</span><span class="p">,</span> <span class="n">b</span><span class="p">):</span>
<span class="k">return</span> <span class="n">m</span><span class="o">*</span><span class="n">t</span> <span class="o">+</span> <span class="n">b</span>
</code></pre></div></div>
<p><img src="/assets/images/2020-12-24-forecast/linear_fit.png" alt="fit data" /></p>
<p>The most naive interpretation of this is to evaluate the <code class="language-plaintext highlighter-rouge">outside_func</code> function at <code class="language-plaintext highlighter-rouge">t=10</code>, which suggests that the value for 2021 will be 31.5 billion USD. However, our goal is not to predict the most likely number, but to produce a distribution. That is, how likely is each specific $y$ value. I think with the fit uncertainty we could analytically build up the probability density function $P(y)$ for the ranges of $y$ we care about. However, I think an equivalent way to do this is to just run a Monte Carlo simulation were we randomly sample from our fit parameters, and then bin the value of <code class="language-plaintext highlighter-rouge">outside_func(t=10)</code> that we get from the simulation.</p>
<p>To generate the Monte Carlo simulation, we want to plot lines that have the parameters <code class="language-plaintext highlighter-rouge">[m,b]</code> drawn from the distribution defined from the covariance matrix of the fit. Wikipedia gives the <a href="https://en.wikipedia.org/wiki/Covariance_matrix#Covariance_matrix_as_a_parameter_of_a_distribution">formula for this joint distribution given the best fit parameters and covariance matrix</a>. This is just the formula for the multivariate normal distribution, which <code class="language-plaintext highlighter-rouge">numpy</code> already has a built in function to sample from. So, we can generate, for example, 1000 samples of the fit parameters in a single line of code:</p>
<div class="language-python highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="n">fit_params_gen</span> <span class="o">=</span> <span class="n">np</span><span class="p">.</span><span class="n">random</span><span class="p">.</span><span class="n">multivariate_normal</span><span class="p">(</span><span class="n">popt</span><span class="p">,</span> <span class="n">pcov</span><span class="p">,</span> <span class="mi">1000</span><span class="p">)</span>
</code></pre></div></div>
<p>where <code class="language-plaintext highlighter-rouge">popt</code> and <code class="language-plaintext highlighter-rouge">pcov</code> are the best fit parameters and covariance matrix returned from the <code class="language-plaintext highlighter-rouge">scipy.optimize.curve_fit()</code> function. Then, we can just plot these 1000 lines, calculate <code class="language-plaintext highlighter-rouge">outside_func(t=10)</code> for each of them, and build a histogram of that distribution in each of the bins selected by Foretell:</p>
<p><img src="/assets/images/2020-12-24-forecast/mc_sim_with_bins.png" alt="binned simulations" /></p>
<p>So, this gives some quantitative probabilities for a starting point. However, notice that this model puts extremely low values on some of the bins. I’m generally skeptical I should ever put 0% in a forecast, and this seems especially to be the case when the model is just curve fitting without outside information. So, my current resolution to this is to combine these values with my initial, uniform prior (which would have put 20% into each bin). I’m maybe 50% confident in this model. So, I will update these values by taking the mean of the uniform prior with the trend-fit value $(0.5 * [\text{prior}] + (1-0.5) * [\text{trend-fit value}])$, and then normalizing the resulting distribution. This gives my predicted values of {10%, 28%, 40%, 12%, 10%}.</p>
<p>So, now we have a sort of outside-view prior to work with. Now the goal is to continue to update these numbers as we gain more information about, e.g., the policies and economies of China and the countries it imports from. That’s beyond the scope of this post, hopefully I’ll write a follow up when I learn any of those details.</p>Christopher PhenicieList of quantum computing companies2020-12-24T00:00:00+00:002020-12-24T00:00:00+00:00https://cphenicie.github.io/list-of-quantum-computing-companies<p>Final update: I won’t be updating this page anymore, <a href="https://en.wikipedia.org/wiki/List_of_companies_involved_in_quantum_computing_or_communication">Wikipedia’s page</a> is probably a better reference</p>
<p>Last content update July 03, 2021. N=31 companies.</p>
<p>I like to keep an updated list of all the companies that are currently trying to build a commercially viable quantum computer. For now, the list only contains the companies that are trying to build the physical quantum computer itself, not including companies that are developing software that runs on a quantum computer (such as <a href="https://qcware.com/">QCWARE</a>) or products to aid in the development of quantum computers (such as <a href="https://q-ctrl.com/products/">Q-CTRL</a>). Instead, this will focus on the companies that actually own the device. I expect that quantum computers will be cloud devices for years to come, so these are the companies that are a candidate to be the AWS of quantum computing.</p>
<p>I have divided up the companies by what type of hardware they are pursuing.</p>
<h2 id="superconducting-qubits">Superconducting qubits</h2>
<ul>
<li><a href="https://www.ibm.com/quantum-computing/">IBM</a></li>
<li><a href="https://research.google/teams/applied-science/quantum/">Google</a></li>
<li><a href="https://www.intel.com/content/www/us/en/research/quantum-computing.html">Intel/QuTech</a></li>
<li><a href="https://damo.alibaba.com/labs/quantum">Alibaba</a></li>
<li><a href="https://www.rigetti.com/">Rigetti</a></li>
<li><a href="https://quantumcircuits.com/">Quantum Circuits</a></li>
<li><a href="https://oxfordquantumcircuits.com/story">Oxford Quantum Circuits</a></li>
<li><a href="https://www.meetiqm.com/">IQM</a></li>
</ul>
<h2 id="trapped-ions">Trapped ions</h2>
<ul>
<li><a href="https://www.honeywell.com/us/en/company/quantum">Honeywell</a></li>
<li><a href="https://ionq.com/">IonQ</a></li>
<li><a href="https://universalquantum.com/">Universal Quantum</a></li>
<li><a href="https://www.aqtion.eu/">Aqtion</a></li>
<li><a href="https://www.oxionics.com/">Oxford Ionics</a></li>
<li><a href="https://www.aqt.eu/">Alpine Quantum Technologies</a></li>
</ul>
<h2 id="neutral-atoms">Neutral atoms</h2>
<ul>
<li><a href="https://www.quera-computing.com/">Quera</a></li>
<li><a href="https://www.atom-computing.com/">Atom Computing</a></li>
<li><a href="https://www.coldquanta.com/#">ColdQuanta</a></li>
</ul>
<h2 id="quantum-dots">Quantum dots</h2>
<ul>
<li><a href="https://quantum.hrl.com/">HRL</a></li>
<li><a href="https://sqc.com.au/">Silicon Quantum Computing</a></li>
<li><a href="https://www.linkedin.com/company/c12-quantum-electronics/">C12 Quantum Electronics</a></li>
<li><a href="https://www.imec-int.com/en/quantum-computing">IMEC</a></li>
<li><a href="https://quantummotion.tech/">Quantum Motion (? They say silicon, not sure if quantum dots or donors in silicon)</a></li>
</ul>
<h2 id="photonics">Photonics</h2>
<ul>
<li><a href="https://www.xanadu.ai/">Xanadu</a></li>
<li><a href="https://psiquantum.com/">PsiQuantum</a></li>
<li><a href="https://www.quix.nl/">Quix</a></li>
<li><a href="https://anametric.com/">Anametric</a></li>
<li><a href="https://lighton.ai/wp-content/uploads/2020/10/White-Paper.pdf">LightOn</a> (Note, this white paper has the only mention of quantum I can find from this company)</li>
</ul>
<h2 id="electrons-on-helium">Electrons on Helium</h2>
<ul>
<li><a href="https://www.eeroq.com/">EeroQ</a></li>
</ul>
<h2 id="topological">Topological</h2>
<ul>
<li><a href="https://cloudblogs.microsoft.com/quantum/2018/06/06/the-microsoft-approach-to-quantum-computing/">Microsoft</a></li>
</ul>
<h2 id="quantum-annealers">Quantum annealers</h2>
<ul>
<li><a href="https://www.dwavesys.com/">D-Wave</a></li>
<li><a href="http://www.qilimanjaro.tech/about/">Qilimanjaro</a></li>
</ul>
<h2 id="companies-that-might-be-building-devices-but-i-cant-tell-from-their-website">Companies that might be building devices, but I can’t tell from their website</h2>
<ul>
<li><a href="http://research.baidu.com/Research_Areas/index-view?id=75">Baidu</a></li>
<li><a href="https://quantum.tencent.com/en-us/">Tencent</a></li>
<li><a href="https://seeqc.com/innovation/dqm-system-on-a-chip/">Seeqc</a></li>
</ul>Christopher PhenicieFinal update: I won’t be updating this page anymore, Wikipedia’s page is probably a better reference