シャドーイング練習: Memory is crazy expensive so I made my own… - 動画で英語スピーキングを学ぶ

There are very few people on the planet who would recognize what this is, and it is incredibly cool.

This is computer memory, not a chip.

There's nothing that's microscopic here, nothing etched in silicon.

Just 64 iron rings, hand threaded and immersed in silicon oil oil.

Each one stores a single bit, and not with electricity, but by manipulating its magnetic field.

This is how computers remembered things long before silicon chips ever existed.

It's the same core technology that allowed the Apollo Guidance computer to fly to the moon, all woven by hand with sewing needles no less.

And that's not the coolest part.

This memory is far more robust than even the stuff that's in your phone today.

It doesn't need power to retain its memory and it can shrug off a radiation burst that would wipe out most chips.

This stuff is nearly indestructible.

And it happens to be a completely insane thing to try to build.

So that's exactly what we're going to do today.

Make our own computer memory from scratch the same way they did 60 years ago.

With some modern tools of course.

And if everything goes well we might be able to make make the world's worst USB drive, or best, depending on how you look at it.

Let's dig in.

The first thing I had to do was to get my hands on some actual memory cores

and then come up with a plan to test them.

Getting a hold of them is easier than you might think.

I got mine off an old Soviet computer, and that matters, and I'll tell you why later.

But for now, I checked in with the boss, made a plan, and got down to research.

And it's kind of crazy, but the best information that I could find came from a 1970s naval manual.

It had fantastic images and really did a good job of explaining all the physics behind it.

And one thing that became clear pretty quickly is that I needed to make a test harness.

And if you watch my channel, you know that I love to hone my skills and learn new ones, So I'm going to be milling out my own circuit boards in-house on my CNC machine.

And it's not as complicated as you might think.

First you use a sharp engraving bit to isolate the copper traces from the rest of the copper.

Then optionally you can apply some solder mask.

I like to because it prevents the copper from oxidizing, but it also makes it a lot easier when you're actually trying to solder later on.

The tricky part about it is removing the solder mask from the areas where you want to see copper, where you're going to solder or have test pads.

This means getting a consistently thin layer so that when you go back with your solder mask removal tool,

you can apply a constant even pressure and it will grind off that solder mask rather than tearing into the copper.

And this is one of those things where you really have to dial it in, test, find different pressures, and expect to mess up.

Alignment is a huge thing too.

If it's off in any way, the tool will start outside the copper and grind into it rather than burnishing it from the top.

But once you find your flow, it's a really cool thing to be able to do.

Now I have some very simple goals for my test harness.

I need to be able to flip a core by passing voltage through two different wires going through it, and then I need to be able to detect that flip on the sense wire.

That's essentially how this all works.

You want to pass enough voltage through a magnetic core in order to flip its polarity.

And when that polarity flips, it induces a current in the sense wire, so the sense wire knows whether a core is flipped or not.

Why does detecting the polarity flip matter?

We'll get to that in a bit.

But for now, we need to finish this up so that we can start writing some firmware for our ESP32.

It's going to help us do those write operations, and we're going to use our oscilloscope in order to detect the rest.

The most basic version of this has one X and one Y wire that supply the current, and then one sense line that's threaded through all of them.

Time to button this up and work on the firmware.

I get a lot of people asking about my background, and it actually started here, in code.

I'm just stubbing out a quick test harness right now to try to see

if we can detect that flip of the single core, but before I ever had a CNC mill, and long before 3D printers were a thing, I was a software engineer.

And look, I'll be honest, I was not a great student.

The traditional sit and read a book or lecture-based learning never really clicked for me.

I was always the hands-on type, so when boot.dev reached out to sponsor the video, I got interested.

One thing that really caught my eye is the gamification.

It's structured like a game.

You're earning XP, leveling up, but most importantly, you're actually doing the work instead of just hearing someone talk about it.

For me, that's the difference between bouncing off of a subject and actually finishing it.

And even though they're pretty backend-focused, they have a lot of different subjects, from Linux to Go to Docker to AWS, real DevOps stuff.

And these aren't just for beginners.

I've been using it to keep my own chops sharp.

I like how you can go at your own pace and skip lessons to what you really need to work on.

The cool thing is all the content is free to read and watch as long as you want, or you can sign up for a membership and that unlocks the interactive part,

the coding, the AI help, and there's a free demo on every course if you just wanna try it out first.

So if you wanna give it a shot, head to boot.dev and use my code for 25% off your whole first year on an annual plan.

The link and the code are in the description, and thanks to boot.dev for sponsoring.

Now let's get back to it.

So this test harness is pretty basic.

It uses an ESP32, a couple of buttons, and a motor controller in order to send current through both of those lines.

I'm hooking up my scope to the trigger of the ESP32 so I know when it fires, and then also on the sense line.

The power from the USB is not enough, and so this requires a separate bench supply in order to deliver the kind of voltage

and amperage needed in order to flip the cores.

My plan is to start out low at 6 volts

and then slowly amp it up until I actually see the flip on the scope.

My trigger looks like it's working well,

but I'm still not getting a flip, so I'll ramp it up a little bit more.

I'm seeing small spikes from the inductive coupling, but I'm not seeing a big dip or spike, so I'm going to try to turn it up to 12 volts.

And after playing around with this for about a half an hour, I see this.

This This dip could indicate a flip.

Now I wrote a zero to it when it was a one, so if I write a zero to it when it's a zero, that dip should go away if it's what I'm looking for.

So I save the waveform and try it again.

I got a bad read on the first one because we don't see any spikes at all, but when I try it again...

That's exactly what we're looking for.

So now it's time to scale this thing up.

My original plan was to do 16x16 cores, giving us a total of 256.

But as I was laying it out, I realized that there's just no way I'm going to be able to fit

that in the form factor of just one board, so I decided to downsize and go 8x8.

So yeah, I guess technically you could call this 64-bit memory.

Kinda.

Once I have an idea of the board size, I go into Shapr and start modeling an enclosure.

The reason I want an enclosure is I want to be able to submerge the entire board in silicone oil.

Now, silicone oil is non-conductive and so it doesn't interfere with the electronics at all, and it's also kind of wild to see a circuit board dunked in liquid.

Its real purpose, other than just looking super cool, is that it helps maintain a consistent temperature throughout all the cores.

And since a core's switching threshold is temperature dependent, it just makes the whole system more robust.

It's far from done, but I feel good about the model and I want to do a fit test.

So it's time to print.

I'm a huge fan of filament printers and I use them all the time in my workflows.

Now I don't think an FDM print is going to be able to contain silicone oil, but it's going to be fantastic for rapid prototyping.

It's an incredible thing to send a model off to a printer

100

and then an hour later have a physical thing to be able to test.

101

When When I'm modeling a physical piece, there's just a lot to keep in my head.

102

The sizes, the tolerances, how it interacts with the things that I'm going to be using it with or putting it in.

103

Oftentimes I miss things, and often it's not until I hold something to where I know

104

that I want to make a design change.

105

Sometimes it's drastic cosmetic changes, and sometimes it's just quality of life stuff when assembling the product.

106

I'm also starting to think through the aesthetic.

107

I want a raw, clean look and big, thick windows that you're going to be able to see through.

108

These windows are going to need to be a precise fit, so I cut up some cast acrylic and move over to my CNC.

109

When I first started working with the CNC, it's less about the machine itself and it's more about knowing the materials.

110

When working with acrylic, for example, you want cast acrylic, not extruded.

111

And you want to cut big chips so that you're not melting the acrylic behind you as you go.

112

Do it right and you can get some awesome results.

113

Ultimately, I want this thing to have a resin enclosure.

114

The print quality is unmatched, but the material selection is far more robust when you're talking about a vessel that's going to contain liquid.

115

I did an initial test print with RIGID 10K by Formlabs.

116

It's got fantastic dimensional stability, it's chemical resistant, and heat resistant as well.

117

So as we let this wash and cure, let's turn our focus back to the production PCBs.

118

These turned out to be a whole lot more difficult than I was expecting.

119

First, I actually needed to make two two-sided PCBs.

120

One for the core memory and the drivers, and the other for the power, the logic, and the sense circuit.

121

I really put my mill to the test on this one, and quite honestly, I know that there's going to be some repairs and some bodges needed afterwards too, but we're going to move forward.

122

By this point, I've got my solder mask application down, but that still doesn't prevent me from screwing things up.

123

They're all just learning opportunities.

124

Right?

125

Homebrew PCBs like this need rivets in the vias.

126

There's no connections through the board unless you make them.

127

So obviously my goal was to avoid using them as much as possible, but yeah, it's just unavoidable.

128

I'm using Kapton tape on the back to avoid the rivets falling out while I try to solder them.

129

is rated up to 260 degrees Celsius, and so it's more than capable of sitting on my hot plate while I solder the rivets in from the other side.

130

Now that I have two PCBs, I need to tie them together using these stacking headers.

131

80 pins total is way more than I need, but it also gives me an opportunity to simplify the design.

132

After the soldering is complete, I clean them up with a little flux cleaner and some IPA and get ready to test the boards.

133

Some of the solder joints need a little cleanup, but other than that, these are looking pretty good.

134

I want to take a second to talk about old PCB design.

135

Hobbyists have been using this method I'm about to show you to hand draw circuit boards since the 50s and 60s.

136

The method is actually really straightforward.

137

It just requires a pen, some copper blanks, and some ferric chloride.

138

And how do you actually draw copper, you ask?

139

Well, you mask it.

140

You draw your design onto the copper using a pen or permanent marker, and then you just let it sit in ferric chloride.

141

Ferric chloride is a copper etchant.

142

It literally dissolves the copper.

143

And you'd think that something that could dissolve copper would eat straight through a permanent marker.

144

Well, luckily you'd be wrong.

145

You let the copper clad fiberglass board sit in the etchin for a while and you'll start to see this black cloud.

146

That is literally the copper being dissolved away from the fiberglass.

147

Keep agitating it and it'll turn completely black.

148

I left it in for 10 minutes, agitating every couple of minutes.

149

Now, I know that this works, but I'd never done it before, and so this was kind of exciting for me too as a test.

150

After the bath, I rinsed it with some clean water, and nothing really happened.

151

But then after I hit it with some isopropyl alcohol, the ink ran off, leaving only the copper behind.

152

Now I'd originally planned on using this technique for the boards, but they were far too complex for me to actually draw something out like this.

153

I was also really surprised to see the detail.

154

Comparing it to this other board from the 70s was really, really cool.

155

I'm going to keep this technique in my back pocket, but for now, let's move on.

156

As I said before, I knew there were going to be changes

157

and bodge wires needed in order to make these boards work, so now's the time to find out just how many and where.

158

I start with basic continuity.

159

I run through the entire board and make sure that all of the things that should be connecting are connected, and some of them weren't.

160

The good news is that the ones that weren't were either low voltage or logic, so they didn't need very big wires.

161

Next, I exported the models from my PCB designer so

162

that I could know exactly how I needed to change my enclosure in order to fit them.

163

The good news is that it looks like the old enclosure can largely fit the board, I just need to make some small modifications.

164

Rather than holding the boards from the side, because I have the edge connectors, I want to actually try to hold them from the middle now.

165

Having the actual PCB models inside Shapr makes things incredibly easy.

166

I can visually tweak and make sure

167

that the holders aren't interfering with any of the components or the traces on the board.

168

And after some test filament prints, it's time to go resin again.

169

Now I get asked a lot whether I prefer FDM or prefer resin prints, and honestly, it's all about the project.

170

I've had some incredible FDM prints that I've sanded and painted that look factory stock, but for resin, it's all about the material properties for me.

171

For most of my functional prints, I do FDM, but when it comes to product prototyping and finished pieces like this?

172

Well, let's put it this way.

173

You can't print pure silicone with FDM yet.

174

After the main enclosure is done, I print two pieces in silicone 40A.

175

Now silicone is a different beast.

176

You can't cross contaminate and it's got its own quirky post-processing steps.

177

For example, using IPA in a sonic bath in order to wash it.

178

Then they need to dry completely before curing.

179

While that happens, we'll finish removing our supports and sanding down the main enclosure.

180

Now Rigid 10K is glass filled, which means that you really need to use protective equipment when sanding.

181

It's quite a bit of work, but it is absolutely worth it.

182

Once the silicone pieces are dried, they need to be cured, but in a water bath.

183

Then it's popped into the cure station to be heated to 60 degrees Celsius and 10 minutes of UV.

184

And while that finishes up, I want to put in my acrylic windows.

185

And you can see me screwing up here by using super glue to bond the acrylic to the case.

186

That was a nice 10 hour detour to reprint the case and then just use the resin itself to set the windows.

187

Now this rigid 10k is hard enough to where I'm able to tap directly into it.

188

I then use a special silicone adhesive called SilPoxy to bond the foot and the top gasket to the case.

189

I'll clean up the excess and put some weights on it while it cures.

190

But this thing is coming together really nice.

191

Now that the case is in good shape and we've repaired the boards,

192

it's time to thread the cores.

193

It's not lost on me that everything that we've done so far is in support of this.

194

Creating this tight interwoven mesh of core memory is really what this project has been about.

195

Now, I deliberately didn't read anything about how to do this.

196

I do that sometimes.

197

I give myself problems to solve that may have already been solved before, but I want to figure it out myself.

198

I start out with oversized wires on the y-axis and then threading eight cores down all of them.

199

And I'm checking the cores along the way because

200

if one is damaged or broken then it's not going to work

201

and I don't want to find out when I've already woven everything together.

202

After that's done, I'll trim the Y wires to size and solder them down.

203

Next, I take the top core from all of those stacks and thread through the X wire.

204

Similar to the Y, I'll oversize it and then trim it and solder it at the end.

205

Now I'm doing two things.

206

I'm checking continuity, but I also want to make sure that I'm not bridging any wires.

207

These wires are enamel-coded, and I want to make sure that I haven't scraped off any of that coating causing a short, because that would kill it too.

208

Once the X's and Y's check out, I thread through the sense wire.

209

The sense wire does just that.

210

It's designed to sense the changing magnetic field of a bit that was 1 transitions to a 0, or a 0 that transitions to a 1.

211

And when it's all done, there's something really beautiful and elegant about the design.

212

So now after all that work, there's only one thing left to do.

213

Let's put it all together and test it out.

214

This is another one of those moments where I know in my head that silicone oil is not conductive,

215

but pouring a liquid onto a live PCB just feels bad.

216

But man does it look cool.

217

Now, I know you're asking yourself, did I have to do this?

218

And the answer is probably not.

219

Now, core memory computers were actually heated back in the day to maintain their coercivity and thus their switching thresholds.

220

But at room temperature and with this small number of cores, I could have probably gotten away with it.

221

I just didn't want to.

222

When I talk about my projects, I always say that it's not about the end result.

223

It's about the journey, and I'm honest about that.

224

I don't always show it in my videos, but my projects tend to take different turns and go down rabbit holes.

225

After all, I usually pick things to work on that I've never done before, which carries a large amount of risk.

226

But it teaches you.

227

It teaches you to think quickly on your feet, to make sacrifices, to make other decisions, accept failures. But this one?

228

This is an example of one that came out amazing.

229

I had originally wanted to use LEDs to show the cores that were lit, but I thought this terminal interface was a lot more appropriate.

230

It allows you to address each core directly, writing a 1 or a 0.

231

And it arranges it all into 8 bytes or 8 characters.

232

Now the ESP32 is doing some heavy lifting behind the scenes, but it exposes a USB drive with a single file called core.txt.

233

And when you write characters to that file, well, it gets set inside those magnetic cores.

234

And when I completely remove power and plug it back in, everything is still there.

235

If you've ever been interested about how a 1 or a 0 translates to something that you can see on the screen,

236

well, this is probably the purest way.

237

And even after you've written characters to it,

238

you can still address those individual cores directly and see how it changes the hex code and the characters.

239

I can honestly say that this is as much fun as I've ever had messing around with 8 bytes.

240

Ever.

241

I gotta say, I've done a lot of cool projects before, but I'm not sure there's been one that's been quite so rewarding, even for something that can only store about eight text characters.

242

And let's talk about just how small that is.

243

This is a USB stick that can hold 128 gigs.

244

It's not the biggest thing out there, but it's not small.

245

That is 16 billion times the storage capacity of these 68 cores.

246

So while core memory has a lot of great attributes, it kind of makes sense why computers used to be the size of entire rooms.

247

And that's just the memory, not to mention the logic and components to access and utilize it.

248

So I hope you enjoyed the video.

249

If you made it this far, please like and subscribe and share and comment and all the things.

250

It really does help YouTube decide to push my channel to a wider audience and helps me keep making videos like this.

251

And before we go, I want to thank our sponsor, Boot.Dev, again for their support.

252

And I'll see you in the next video.

✨ おすすめ動画

192 文

この動画で鍛える英語スピーキングの目標

この動画は、英語スピーキング練習において「技術的な説明を流暢に話す」というマイルストーンを目指します。専門用語や工程の説明を含む長文を自然に発音し、リズムを保つ能力を養います。特にIELTSスピーキング対策にも有効で、複雑な内容を明確に伝えるスキルが向上します。

実用フレーズバンク

「It's the same core technology that allowed...」（...を可能にしたのと同じ核心技術です）- 技術の共通性を説明する際に使えます。
「The tricky part about it is...」（その難しい点は...です）- 課題のポイントを指摘する表現。
「Once you find your flow, it's a really cool thing to be able to do」（リズムを見つけると、とても楽しいことです）- 習熟した際の感想を伝えるフレーズ。
「I need to be able to flip a core by...」（...によってコアを反転させることができなければなりません）- 目的を達成するための手段を説明する文型。

弱点を克服するポイント

この動画は英語シャドーイング（shadowing）に最適で、「子音の発音とリズムの維持」を鍛えます。特に「microscopic」「manipulating」などの長い単語では子音のクリアさが重要で、「and that's not the coolest part」のような接続詞のリズムが全体の流れを決めます。shadow speechの練習では、動画のスピーカーに合わせて発音のタイミングを合わせることで、自然なリズムを身につけられます。また、「solder mask」「circuit boards」などの複合語のアクセント位置も注意が必要で、繰り返し練習することで正確な発音ができるようになります。

これらのポイントを意識してshadowing siteを利用して練習すれば、技術的な説明も自信を持って話せるようになります。

この動画の文法

話し手がよく使っている文型を、動画の実際の表現とともに紹介します。

文型	動画での表現
現在完了進行形 have/has been + -ing — 以前に始まり今も続いている動作	I've been using · have been using
受動態 be + 過去分詞 — 誰がするかより、何が起きるかに焦点を当てる	is flipped · is unmatched · is rated
現在完了形 have/has + 過去分詞 — 過去の出来事が今も関係している	I've sanded · we've repaired · we've done

シャドーイングとは？英語上達に効果的な理由

シャドーイング（Shadowing）は、もともとプロの通訳者養成プログラムで開発された言語学習法で、多言語習得者として知られるDr. Alexander Arguelles によって広く普及されました。方法はシンプルですが非常に効果的：ネイティブスピーカーの英語を聞きながら、1〜2秒の遅延で声に出してすぐに繰り返す——まるで「影（shadow）」のように話者を追いかけます。文法ドリルや受動的なリスニングと異なり、シャドーイングは脳と口の筋肉が同時にリアルタイムで英語を処理・再現することを強制します。研究により、発音精度、抑揚、リズム、連音、リスニング力、そして会話の流暢さが大幅に向上することが確認されています。IELTSスピーキング対策や自然な英語コミュニケーションを目指す方に特におすすめです。

シャドーイングのやり方: ステップ別の完全ガイドを読む →

シャドーイング練習: Memory is crazy expensive so I made my own… - 動画で英語スピーキングを学ぶ

Making RAM at Home

If I started a YouTube channel from scratch, I’d do this

THIS Is Where the Internet Lives

what chess does to your brain

I Build a Giant NASA SLS Water rocket

What Can a $50 Laptop Do?!? (2026)

25 crazy software bugs explained

Making a 360° Controller Smaller than a Penny!

How I use AI to learn engineering (anti brain rot studying)

9 Investments for Better Learning

Building the world's largest WORKING iPhone!

Building an Editing PC Right Now? Yikes.

この動画で鍛える英語スピーキングの目標

実用フレーズバンク

弱点を克服するポイント

この動画の文法

シャドーイングとは？英語上達に効果的な理由